Aller au contenu
Corporate & Investment Banking Mon parcours English

Accueil / Capital Markets & Securities Services / Capital Markets / Module 6

6Sales & Distribution

Connecting Markets Products to the Investors Who Buy Them · The People, Platforms, and Economics of Distribution

Les équipes et plateformes qui amènent les produits jusqu’aux investisseurs.

Lecture
2 h 30
Sections
69
Questions de rappel
37
Questions de QCM
20

Le texte de ce module est encore en anglais : la version française arrive. Les outils interactifs, eux, sont déjà en français.

Avec qui ce métier travaille
les ordres des investisseursEquity Capital Markets (ECM) : les ordres des investisseurs1ECMplacement des obligationsDebt Capital Markets (DCM) : placement des obligations2DCMdistribution aux investisseursStructured Products : distribution aux investisseurs4Structured Productsclient autorisé à traiterClient Onboarding and KYC in Global Markets : client autorisé à traiter5KYCordres actionsEquity Trading (Cash & Algo) : ordres actions9Equity Tradingordres obligatairesFixed Income Trading : ordres obligataires10Fixed Incomeordres de changeFX Trading (Spot, Forwards, Swaps) : ordres de change11FXidées et notes pour la venteResearch & Investment Advisory : idées et notes pour la vente7Researchdemandes des clientsCommodities Trading : demandes des clients12Commoditiesidentifiants des investisseursMarket Data & Reference Data : identifiants des investisseurs28Market & Ref Databesoins des clientsDerivatives Structuring (OTC) : besoins des clients3Structuringopérations clientsDerivatives Trading : opérations clients13Derivatives Tradinglimites de contrepartieCounterparty Risk : limites de contrepartie27Counterparty Riskdemandes des fondsPrime Brokerage & Client Services : demandes des fonds8Prime Brokerage6

Distribution is the bridge between what the bank makes and what the investor holds

Every other capital-markets module ends where this one begins. A new bond is priced and allocated; a swap is structured; a block of shares is underwritten. None of that matters until the instrument is placed with an investor and can be traded again tomorrow. Sales and distribution is the client-facing layer that sits across the entire markets business — the membrane every product must pass through to reach the buy-side. It is simultaneously the bank's revenue engine, its deepest competitive moat, and, over the last decade, the part of the franchise most violently reshaped by technology and regulation.

The reader who finishes this module will be able to read a distribution franchise end to end. They will know who the investors are and why a salesperson treats each type differently, how a trade travels from a market idea to a settled position, why the same flow that once moved by telephone now moves by platform and API, and where the money actually comes from once electronification and MiFID II have finished compressing it. They will also see why almost every pain point in the business turns out to be a data and integration problem — which is exactly why it is a technology story as much as a sales one.

Two threads run through everything: the business and the technology

The business thread is a competition for wallet. Each investor spends a finite pool of fees, spreads, and commissions across all the banks it deals with. That pool is its wallet, and the sell-side competes for a share of it with better prices, better ideas, better access, and better service. Distribution is the machinery of that competition. Read the business thread and you understand why relationships, information, and content are the real currency of the sales floor, and why a bank will fight for a top-three position in a client's wallet the way it fights for a place on a league table.

The technology thread is the migration of that competition onto platforms. Flow that a salesperson once quoted by voice now streams from a pricing engine, thousands of times a second, into venues where a client puts five banks in competition with one click. As the flow electronifies, the edge shifts from relationship to price, speed, and data — and distribution becomes part relationship business, part software product. Read the technology thread and you understand why connectivity, CRM, e-pricing, and surveillance are where the modernisation budget goes, and why "unify the client data" is the sentence behind nearly every transformation programme in the client-facing business.

How to read this module

Read it as a narrative, not a reference card

The parts build on each other and are meant to be read in order. Part 1 sets up what distribution is and who it serves. Part 2 walks a trade from idea to settlement and shows where the process has gone electronic. Part 3 names every actor and team. Part 4 opens up the systems and data underneath. Part 5 turns to what goes wrong, how to measure quality, and where the model is heading. A single fictional bank, a fictional issuer, and two fictional investors recur throughout so that each mechanic lands on a concrete example rather than an abstraction. Every figure attached to them is illustrative and flagged as such.

Each coloured box plays a fixed role

The boxes are not decoration; each colour means one thing. A blue Key insight box crystallises an idea worth remembering. A navy Going deeper box adds detail a specialist would want and a generalist can skip on first read. A teal IT & data lens box names the systems and data that carry the mechanic just described. An orange Watch out box flags a trap or a live area of change. A green Worked example box follows the running case. A light-blue In a nutshell box recaps a part, and a purple Check your understanding box closes it with questions and answers.

What you will master

You will understand the distribution model end to end

By the close you will be able to trace a product from the desk to the investor and back. Primary placement and secondary flow, high-touch voice and low-touch API, vanilla and bespoke — you will know which path a given product and client take, and why.

You will be able to map every actor

You will know who sits where and who talks to whom. Coverage sales, specialist sales, sales-trading, syndicate, e-commerce, the desk, research behind its wall, compliance, onboarding, and the venues and vendors in between — and why the salesperson sits in the middle of all of it.

You will understand where the price actually comes from now

You will see that for liquid flow the human has left the pricing loop. A pricing engine computes a mid, applies a spread, skews for the desk's axe, checks the limits, and answers in milliseconds. You will understand the inputs, the risks, and what the salesperson does once the machine quotes.

You will be able to trace data front to back

You will follow a client interaction from CRM through pricing, execution, and booking to transaction reporting and surveillance. You will see why fragmented client data is the core weakness of the whole model, and why the "360-degree view" is the prize everyone is chasing.

You will recognise the risk and transformation landscape

You will know how distribution fails and how good distribution is measured. Six risk types, a handful of KPIs, three real scandals that each expose a different failure mode, and the transformation trends — electronification, data and AI — that are rebuilding the model, assessed honestly rather than sold.

1Context and the Big Picture

Before the platforms and the protocols, one idea: a bank makes prices and products, and investors have money to put to work. Distribution is everything that connects the two. This part builds that connection from zero — what the business is, where it sits, what it distributes, who buys, and how it makes money — and introduces the running cast we will follow for the rest of the module.

Distribution turns products and liquidity into revenue

Start with the simplest possible version

A bank's markets business produces two things an investor might want: products and liquidity. A product is an instrument — a bond, a share, a swap, a structured note. Liquidity is the ability to buy or sell one of those instruments now, at a fair price, in the size you need. The trading desks and the new-issue teams manufacture both. But manufacturing is worthless on its own. Someone has to find the investor for whom this bond, this swap, this price is the right thing today, reach that investor through some channel, agree a price, and get the trade done. That someone is sales, and the whole apparatus around them is distribution.

Distribution, defined precisely, is the process and the network through which a bank makes its markets products and prices available to clients and executes their business. Sales is the set of people who own the client relationships and drive that flow — surfacing what clients need, sharing ideas and prices, and getting trades done. The "model" is how the bank organises all of this — by client, by product, by channel — to cover the market efficiently and profitably. Strip away the jargon and the job is a bridge: it carries the bank's inventory to the investor's portfolio, and carries the investor's orders back to the desk.

The business answers three questions every single day. Who are the right investors for this product or this flow? How do we reach them — a call, a chat, a platform, a portal, an API? At what price, and at what cost to serve, do we transact and support them? Everything in this module is ultimately an elaboration of those three questions. The first is a problem of client knowledge and data. The second is a problem of channels and technology. The third is a problem of economics.

Sales faces two directions at once, and that is the whole job

Facing inward, sales represents the client to the desk. The salesperson walks to the trader and says, in effect, "here is what my accounts need, here is where real demand sits, here is the price sensitivity I am seeing." That intelligence is priceless to a desk deciding where to show risk and how to price it. Facing outward, sales represents the bank to the client: "here is our price, here is an idea worth your time, here is where we are axed." The salesperson is a trusted intermediary standing between two parties who each know something the other needs, not an order-taker relaying instructions. That two-way role — and the information asymmetry it manages — is why relationships and trust carry so much economic weight in this business, and why so much of the module returns to who is allowed to know what.

Distribution is the client-facing membrane of the whole markets business

It sits across everything, not beside it

In the markets value chain, distribution is not one link; it is the surface. Upstream, the new-issue teams in equity and debt capital markets create products, and the trading desks create liquidity. Structuring designs bespoke solutions for clients with specific needs. All of that output has to reach an investor, and the only path to the investor runs through sales. Downstream, once a trade is agreed, execution and post-trade take over — the trade is confirmed, cleared, and settled. Distribution is the client-facing membrane in the middle: everything the bank makes passes through it on the way out, and every order the client sends passes through it on the way back.

This position is why distribution is strategically central rather than a back-office afterthought. It monetises the entire franchise — client flow drives commissions, spreads, and new-issue fees across every product the bank makes. The relationships it owns are the bank's real moat, hard to build and increasingly powered by data. It is being reshaped in real time by electronification, which is a technology story before it is a commercial one. And it is heavily regulated as a matter of conduct: how a bank distributes is now a compliance discipline in its own right, not a courtesy it extends to clients. A bank that neglects distribution is a factory with no loading dock.

Two modes of distribution: placing the new, and trading the flow

Primary and secondary are different games played by the same people

Primary distribution is the placing of brand-new issues. When a company or a government issues new bonds or shares, the bank's syndicate and sales force place them into investor books at launch. It is a one-off, time-boxed, syndicate-led process: canvass investors for demand, build a book of orders that sets the price, then allocate the — usually oversubscribed — issue across accounts. It is intense and lasts days, and it earns a fee or selling concession from the issuer. This is where the equity and debt capital-markets modules hand the baton to distribution: sales is the distribution muscle behind every new issue.

Secondary distribution is the ongoing, everyday flow. It is the continuous business of quoting prices, providing liquidity, and executing investors' buy and sell orders in existing instruments — all day, every day. The desk either takes the other side of the trade itself, as principal, or works the client's order into the market as agent. It earns a spread or a commission per trade. Most standard secondary flow now runs on electronic platforms.

The two modes are joined at the hip, and the join is the reason primary works at all. An investor buys a new bond in primary only because it believes it can sell that bond later in secondary. Secondary liquidity is the promise that makes primary demand possible. A bank that is a strong secondary market-maker in a name is a more credible primary distributor of that name, because its investors trust that the aftermarket will be there. This is the first appearance of a pattern that recurs throughout the module: the parts of distribution are not independent; they reinforce each other, and a weakness in one quietly undermines the others.

Distribution spans a spectrum from vanilla flow to bespoke structure

Sort the products by liquidity and margin, and the model becomes legible

Everything the markets business makes can be laid on a single axis, and that axis governs how it is distributed. At one end sit vanilla, high-volume, low-margin, electronic products — cash equities, foreign-exchange spot. At the other sit bespoke, low-volume, high-margin, relationship-and-advisory products — structured notes, complex derivatives. The distribution model — which channel, how much automation, how intensely a human covers it — is calibrated to where a product sits on that axis. Push standard flow toward automation; reserve human coverage for where it adds real value. The table below sorts the main families and shows how the model changes as you move along the spectrum.

Product family What it is Distribution style Margin Key technology
Cash equities Shares and ETFs Agency, high-volume, heavily electronic; sales-trading central Thin commission Order/execution management, algos, exchange connectivity
Rates (FICC) Government bonds, swaps, futures Principal liquidity; RFQ and streaming, rising electronic share Spread, compressing Pricing engines, MDP connectivity, FIX
Credit (FICC) Corporate bonds, CDS Relationship-heavy, fastest-electronifying; portfolio trading Spread, still meaningful Axe-matching analytics, MarketAxess/Tradeweb, e-credit pricing
FX (FICC) Spot, forwards, options The most electronic market of all — streaming, algos, API Very thin spread Low-latency pricing, streaming, API distribution
Listed & OTC derivatives Futures, options, swaps Cleared flow on venues; bespoke structures stay high-touch Mixed SEF/OTF connectivity, structuring and pricing tools
Structured products & financing Bespoke notes, repo, securities financing Advisory-led, suitability-heavy, largely human High, embedded Issuance/pricing platforms, suitability and KID tooling

The spectrum is not static; products migrate leftward along it over time. Credit is the live example. A decade ago corporate bonds were a voice business because they were illiquid and heterogeneous — thousands of individual line items, many trading rarely. Two protocol innovations, portfolio trading and all-to-all, changed the economics of trading them electronically, and credit is now the fastest-electronifying market in the building. The lesson for anyone assessing a distribution business is that the boundary between "human" and "automated" flow is a moving front, not a fixed line, and the technology that moves it — a new protocol, a better pricing model, cleaner data — is where competitive advantage is won and lost.

FICC is the larger distribution engine, and it is built on principal risk

The markets business splits in two, and the halves distribute differently

The markets business divides into two great halves — FICC and equities — and distribution works differently in each. FICC stands for fixed income, currencies, and commodities: it houses rates, credit, foreign exchange, and commodities, and it is the larger revenue pool at most banks. The equities complex houses cash equities, equity derivatives, and the financing business — prime brokerage and securities lending — that serves hedge funds. The distinction is not administrative; it goes to how money is made. FICC is predominantly a principal business: the desk quotes a two-way price, takes the other side onto its own book, and earns the bid-offer spread for warehousing the risk. Equities is more of an agency business: the bank works the client's order into the market and earns a commission without taking the position. That single difference — principal risk versus agency commission — shapes everything downstream, from how the flow is priced to how much balance sheet the business consumes to which regulations bite hardest.

Figure 1 Illustrative relative revenue pools across the markets business. FICC — rates, credit, FX, commodities — is the larger, more principal-driven distribution engine; the equities complex is more agency and financing. Orders of magnitude only.

Because FICC takes principal risk, its distribution model is inseparable from the balance sheet, and that is a strategic fact. When the bank quotes Meridian a price on a corporate bond and takes the bond onto its book, it is committing capital and consuming balance sheet until it can offload the position — so FICC distribution is rationed by risk appetite and capital in a way agency equities is not. This is why the axe matters so much in FICC: the desk is not just serving the client, it is managing an inventory it is legally on the hook for. It is why FICC pricing engines must check risk and limits at machine speed. And it is why the post-crisis capital rules, which made warehousing risk more expensive, pushed FICC toward agency-like electronic protocols such as all-to-all, where the bank can match two clients without taking the risk itself. The reader who remembers one thing about FICC should remember that its distribution is a balance-sheet business wearing a sales suit.

FICC and equities distribution, compared

The two halves distribute so differently that laying them side by side sharpens the whole picture. The table contrasts them on the dimensions that matter — how the bank is paid, whether it takes risk, how electronic the flow is, and what the technology has to do. Read it as a reminder that "distribution" is not one operating model but a family of them, each calibrated to how its products trade.

Dimension FICC distribution Equities distribution
Primary revenue Bid-offer spread (principal) Commission (agency)
Risk taken Yes — warehouses inventory Mostly no — works the order
Balance-sheet intensity High Low
Dominant venues RFQ / streaming MDPs, dealer-to-client Exchanges, CLOB, dark pools
Human role Pricing exceptions, illiquid, structured Blocks, capital commitment, complex orders
Key technology Pricing engine, axe-matching, connectivity Algos, smart order router, DMA, EMS

Meet the cast: the running example for the whole module

One issuer, one deal, two investors — all illustrative

A single set of fictional characters will make every later mechanic concrete. They are invented, and every number attached to them is illustrative only, chosen to teach rather than to report. Introducing them now means that when Part 2 dissects bookbuilding or Part 4 traces a data lineage, the example is already familiar.

The bank is a mid-sized European corporate-and-investment bank with a full markets franchise — a rates desk, a credit desk, an FX desk, an equities business, a syndicate, and a growing e-commerce team. It is strong in euro credit and euro rates, competitive but not dominant in FX, and investing hard to defend its position as flow moves electronic. It is our vantage point throughout.

The issuer is Lumen Energy SA, an investment-grade European utility that needs to raise seven-year money. Its treasury team will mandate the bank, among others, as a joint bookrunner on a EUR 750 million senior benchmark bond. Lumen is the primary-distribution thread: we will follow that bond from mandate to allocation to aftermarket.

The two investors sit at opposite ends of the buy-side. Meridian Asset Management is a large, benchmark-driven, buy-and-hold real-money fund — the backbone of a bond book and a top-tier relationship the bank covers across rates, credit, and FX. Kestrel Capital is a fast-trading, leveraged multi-strategy hedge fund — a high-value, sophisticated, price-sensitive account that lives on speed, ideas, and electronic execution. Meridian is the relationship thread; Kestrel is the electronic-flow thread. Between the two, they carry almost every point this module needs to make.

How distribution makes money, and why the economics are under pressure

Four revenue mechanics, two of them reshaped by regulation

Distribution earns its keep four ways, and knowing which is which explains a lot of behaviour. On principal trades the desk buys at the bid and sells at the offer, and the bid-offer spread is its reward for providing liquidity and warehousing risk — the core of FICC and derivatives revenue. On agency business, especially cash equities, the bank executes the client's order and charges a commission without taking the risk onto its book. In primary distribution the issuer pays a fee, shared among the syndicate banks, for placing the deal — the reward for distribution reach. And on bespoke or structured products the margin is embedded in the price, higher because it reflects design, risk, and advisory work.

Revenue mechanic Where it applies What the bank is paid for Direction of travel
Bid-offer spread Principal trades (FICC, derivatives) Providing liquidity, warehousing risk Compressing with electronification
Commission Agency trades (cash equities) Executing without taking risk Thin and competitive
New-issue fee / concession Primary distribution Distribution reach into investor books Stable but shared across syndicate
Structuring margin Bespoke / structured products Design, risk, advisory Highest; the defensible end

Two of these four were reshaped by MiFID II, the European rulebook that governs how banks distribute. The first change was research unbundling: banks could no longer bundle research into trading commissions as a free inducement to trade, so content had to be priced and paid for as a separate product. The second was a broader curb on inducements plus a duty to disclose all-in costs. Combined with electronification compressing spreads and commissions, these reforms put distribution economics under structural pressure — which is precisely what raises the premium on efficiency, data, and higher-margin advisory business. When margin is falling, cost-to-serve becomes the battleground.

Wallet share is the league table of distribution

Each investor spends a finite pool across all its banks, and that pool is the scoreboard. A client's total trading-and-fee spend — its wallet — is split among the banks that serve it, and every bank wants the biggest slice of the most valuable wallets. Wallet share plays the role that league tables play in new issuance: it is the number that quietly shapes behaviour. It is why a bank will do a low-margin trade to stay top-of-mind, why coverage teams obsess over "broker votes" and client scorecards, and why losing a place in a top client's top three is treated as a strategic wound. The competition is not for any single trade; it is for a standing share of a relationship, and distribution is the machinery that competes for it.

How the wallet is actually paid, and why the plumbing shapes behaviour

Wallet share is an abstraction until you see the concrete ways a client pays it, and the mechanics themselves shape behaviour. In agency equities, a large buy-side client often runs a commission-sharing arrangement: it directs execution commissions to the banks that execute its trades, then instructs a portion of those commissions to be paid on to the research providers it values, whether or not they did the trade. This splits the payment for execution from the payment for research even inside the commission — a precursor to what MiFID II later forced fully into the open. In FICC there is no explicit commission; the client pays through the spread it accepts, so the wallet is embedded in prices rather than invoiced, which makes it harder to measure and is exactly why a bank invests so heavily in analytics to estimate what each client is really worth. The broker vote is the periodic ritual that ties it together: the buy-side scores its providers on research, execution, ideas, and service, and those scores drive how the wallet is allocated for the next period.

The practical consequence is that a salesperson is always managing two currencies at once — the trade and the vote. A trade that loses a little money can still be worth doing if it protects a relationship that delivers a strong broker vote and a larger wallet next quarter. A brilliant piece of research that wins the analyst a top vote justifies its cost even though research no longer trades for commission directly. The distribution model exists to optimise this two-currency system across hundreds of clients — which trades to do, which content to give away, which relationships to defend — and it is why the coverage salesperson thinks less like a shopkeeper closing sales and more like a portfolio manager allocating scarce attention and balance sheet to the relationships with the best long-run return.

A few reference numbers make the business concrete

Distribution is a large-revenue, thin-margin business, and a handful of illustrative figures fix the scale. The global markets divisions of the major banks earn tens of billions of dollars a year between them from sales and trading, split roughly between the equities complex and the larger fixed-income, currencies, and commodities complex — FICC is the bigger distribution pool. Within it, a single top-tier institutional client might pay a bank tens of millions a year across all products in a strong year, spread across dozens of legal entities and thousands of trades. A benchmark bond deal like Lumen's carries a distribution fee measured in tenths of a percent of the amount raised, shared across a syndicate of several banks — thin per deal, but multiplied across a heavy calendar. And the electronic share of that flow keeps rising, so the revenue per trade keeps falling even as the number of trades climbs. The numbers that matter are therefore not absolute revenue but the ratios: wallet share, revenue per client, and cost-to-serve, because in a thin-margin business it is the ratios, not the totals, that decide who is winning. Every figure here is an illustrative order of magnitude, not a quoted fact.

Index inclusion quietly drives a large share of bond demand, and it is worth naming. A great deal of real-money buying is benchmark-driven: funds that track or are measured against a bond index must hold, in roughly the right proportion, the bonds that qualify for that index. When Lumen's bond meets the criteria for the main euro investment-grade indices, a wave of index-following demand appears almost mechanically, independent of any view on the credit — and when a bond is too small or too short to qualify, that demand simply is not there. A distributor who understands which accounts are benchmark-driven, and whether a given bond will be index-eligible, is reading a large part of the demand curve before the book even opens.

The relationship is the real moat, and it is scored like a league table

A price can be matched overnight, but a relationship takes years to build, and that asymmetry is the franchise. Any competitor can quote the same spread on the same bond; what it cannot instantly replicate is a decade of trust, the salesperson a client calls first, and the accumulated knowledge of what that client needs before the client asks. This is why banks treat the relationship, not any individual trade, as the asset — and why they measure it obsessively. Clients periodically score their banks in "broker votes" or counterparty reviews, ranking each provider on price, ideas, service, and access; those scores drive the client's allocation of its wallet, and a bank's coverage teams live and die by them. The scoring closes a loop: better content and service earn a better vote, a better vote earns more wallet, and more wallet justifies more coverage and technology invested in the relationship.

The economics of this are worth stating as a value equation, because it drives every coverage decision. Profitability on a client is its revenue — wallet share times client value — minus the cost of serving it. Some clients pay more than they cost and are worth intense human coverage; some cost more than they pay and should be served cheaply through electronic channels; and the whole point of the distribution model is to maximise this equation across the client base. Tier the clients, concentrate scarce human coverage and balance sheet on the most valuable, automate the long tail where cost-to-serve approaches zero, price by value so that clients who bring more wallet get better terms, and measure everything. That value equation is the quiet logic behind the org chart, the channel strategy, and the technology budget alike, and it reappears in every later part of this module.

The business is cyclical, and two cycles set its tempo

Volumes rise and fall with markets, but the deeper cycle is technological

Distribution volume tracks market activity, so revenue is cyclical in the obvious way. When markets are volatile and issuance is heavy, flow surges and spreads widen; when markets are quiet, flow thins and competition for it sharpens. That is the short cycle, and every distribution business rides it. But the more consequential cycle for anyone building or assessing the franchise is structural, not seasonal: the steady, one-directional migration of flow from voice to platform to API. Unlike the market cycle, electronification does not reverse. Each asset class electronifies at its own pace — FX and equities long ago, rates well advanced, credit accelerating now, structured products barely started — but the direction is always the same, and it is set by liquidity and standardisation.

The two cycles interact in a way that matters for technology investment. A downturn in flow does not pause electronification; if anything it accelerates it, because falling revenue makes cost-to-serve urgent and automation is how cost comes out. And a regulatory cycle sits on top of both — MiFID II pushed flow onto venues deliberately, and each new rule adds a data and reporting obligation that only technology can carry at scale. So the distribution business a bank runs in any given year is shaped by where it sits in three cycles at once: the market cycle, the electronification curve, and the regulatory calendar. Read those three together and you can predict where the pressure — and the investment — will land.

The recent cycle tells the story cleanly, and it is a rate story

The last few years show the distribution cycle at work, and the driver was rates, not equities. Run the illustrative arc. Through 2020 and 2021, with policy rates near zero and central banks buying bonds, credit spreads were tight, issuance was heavy, and real-money demand was reliable; distribution was a high-volume, tight-margin business where the game was reach and efficiency. Then in 2022 and into 2023, as inflation forced rates sharply higher, volatility spiked, spreads widened, and the picture inverted: bid-offer spreads widened as liquidity became scarce and valuable, some issuance windows slammed shut, and the desks that could provide liquidity in a stressed market earned well for it. By 2024, with rates elevated but more stable, a new equilibrium settled in — higher all-in yields that drew income investors back into bonds, but a more discerning market that punished weak credits. The point of the arc is that FICC distribution breathes with the rate cycle: volatility and dispersion are, up to a point, good for a liquidity provider, while the quiet, everything-tight markets that look benign are actually the hardest place to earn a spread.

This is why the cycle governs technology investment as much as revenue, and why the two cycles interact. In the tight-margin phase, the pressure is to automate — to serve more flow more cheaply, because margin per trade is thin — so electronic build accelerates. In the volatile phase, the pressure is on risk and resilience — pricing engines must not misfire when volatility spikes, connectivity must not drop when everyone trades at once — so the investment tilts toward robustness. A bank reading its own three cycles at once — the rate cycle setting volatility and spreads, the electronification curve setting the direction of build, and the regulatory calendar setting the compliance load — can predict where next year's distribution pressure, and next year's technology budget, will land. That foresight is worth more than any single trend, because it tells the business where to spend before the pressure arrives.

  1. What two things does a bank's markets business manufacture, and why is manufacturing worthless without distribution?

  2. Distinguish primary from secondary distribution, and explain why one depends on the other.

  3. What single axis organises the product spectrum, and how does the distribution model change along it?

  4. Why is FICC distribution described as a balance-sheet business, and how does that differ from equities?

  5. What is wallet share, how is it actually paid, and why does it function like a league table?

  6. Why is electronification a more consequential cycle than the market cycle for technology investment?

1

  1. Why is manufacturing worthless without distribution?

  2. Why do investors buy in the primary market?

  3. Along which axis does the distribution model change from electronic to high-touch?

  4. Why is FICC distribution a balance-sheet business?

2The End-to-End Flow

This is the longest part of the module, and the most important. It follows a trade from the moment an idea forms to the moment it settles and the relationship rolls on — seven steps that every piece of distributed flow passes through. Along the way it stops at the mechanics that matter most: the channels that reach the client, the protocols that get the trade done, the platforms where banks compete, the pricing engine that has quietly replaced the human quote, and the bookbuilding and allocation that place a new issue. Watch for where the process has gone electronic, because that is where the business is being rebuilt.

Hold the whole arc in your head before the detail

A trade travels seven steps from idea to settled position

Every distributed trade, however it is executed, moves through the same seven steps. Sales absorbs content and forms a view worth sharing. It reaches out to the right clients with that view. A client asks for a price or expresses interest. Sales prices and negotiates. The trade is executed and confirmed. For blocks and new issues it is allocated across accounts, then it clears and settles. Finally the relationship is serviced — reporting, best-execution evidence, after-sales support, and the next idea. The steps are the same whether the trade is a EUR 5 million government bond that a machine prices in milliseconds or a bespoke structured note that takes three weeks of phone calls. What changes is how much of each step a human does and how much a platform does.

Step What happens Who leads Where it has gone electronic
1. Content & idea Absorb research, colour, and the desk's axes; form a view Sales, research, desk Data-driven idea generation, next-best-action
2. Reach out / pitch Contact the right clients with a relevant idea or price Sales, e-commerce Axe broadcast, targeted analytics
3. Enquiry / RFQ Client asks for a price or shows interest Client RFQ on platforms; the workhorse of e-trading
4. Price & negotiate Get a price from the desk or the engine; quote the client Sales, desk, pricing engine Auto-pricing; the human is out of the loop for liquid flow
5. Execute & confirm The trade is done, captured, and confirmed Sales-trading, desk Click-to-trade, auto-execution, straight-through capture
6. Allocate & settle Allocate blocks/new issues; clear and settle Syndicate, ops Electronic bookbuilding tools; STP to settlement
7. Service & follow-up Reporting, best-ex evidence, support, the next idea Sales, controls Automated reporting, TCA, analytics

The pressure is not spread evenly across the seven steps

Effort and risk concentrate in the middle of the arc, at enquiry, pricing, and execution. The first two steps are preparation — absorbing content and reaching out — and they are continuous and relatively low-stakes on any single trade. The last two, allocation-and-settlement and service, are consequential but largely mechanical once the trade is agreed. The intensity spikes at steps three, four, and five: the moment a client asks for a price, the bank has seconds to a few minutes to compute the right number, check its risk, decide whether to win the trade and at what level, and execute cleanly. Get the price too wide and the trade goes to a competitor; too tight and the bank loses money; too slow and the opportunity is gone. That is the peak of the curve, and it is exactly the part of the process that electronic platforms have automated most aggressively.

Figure 2 The distribution workflow: effort and risk peak at the enquiry-price-execute core, which is also where electronification has advanced furthest. Illustrative.

Step 1 — A view worth sharing is the price of the client's attention

Sales runs on differentiated content, and content is now a costed product

A salesperson earns the right to reach a client by making that client smarter. The raw material is research — analysts' views on companies, sectors, economies, and rates — plus strategy and market colour on what is moving flows, trade ideas that turn a view into an actionable, sized, priced position, and the desk's axes and liquidity information. The best content is differentiated: something the client cannot get from every other bank. This is the fuel for the whole workflow, because a pitch built on a genuinely useful view gets returned; a pitch built on nothing gets ignored.

Two facts complicate step 1, and both are structural. The first is the information barrier: research is separated from sales and trading by independence rules, so sales distributes research but cannot direct it, and the private-side deal teams cannot feed the public-side sales floor. The second is that research is no longer free. Since MiFID II, content is a priced product with a budget attached rather than a giveaway bundled into commissions — a rule that has partly reversed under the 2024 Listing Act but still governs how European banks cost and distribute what they know. The consequence for the workflow is that the content feeding step 1 is scarcer and more valuable than it was, which raises the premium on generating ideas from the bank's own flow and data rather than from research alone.

Step 1 is increasingly a data problem. The modern version of "form a view worth sharing" is next-best-action: a model reads a client's history — every trade, RFQ, and research interaction — alongside live market data and suggests the most relevant idea, axe, or product for that specific client today. It runs on a unified client dataset, the "360-degree view" that Part 4 dissects, and its output is only as good as that data is clean. The unglamorous truth is that most of the work in data-driven idea generation is not the model; it is joining fragmented client data into something a model can read.

Step 2 — Reaching the client means choosing a channel, and the channel is the strategy

Distribution flows through a spectrum from a phone call to an API

The bank reaches clients through a spectrum of channels, and each sits at a different point on a single trade-off. At the high-touch end is voice — phone calls, relationship-rich, slow but personal, reserved for complex, large, or illiquid trades and ideas. Next is chat and instant messaging over Bloomberg, Symphony, or Teams: the everyday workhorse, fast, logged, and semi-structured. Then the bank's own single-dealer platform, a branded portal carrying prices, axes, research, and execution. Then third-party multi-dealer platforms, where the client puts several banks in competition on one trade. At the low-touch end is direct API and algorithmic trading, fully automated, the client's system trading against the bank's prices with no human in the loop.

Figure 3 The channel spectrum: from high-touch voice, where margin and human value are highest and liquidity lowest, to low-touch API, where flow is automated, standardised, and thin-margin. Each product-and-client combination is assigned to the cheapest channel that still serves it well. Illustrative.

The organising principle is to assign each product-and-client combination to the cheapest channel that still serves it well. A bespoke structured note needs a phone call; a EUR 5 million government-bond trade for a systematic account goes on a multi-dealer platform; FX spot for a high-frequency fund streams over an API. As you move left to right across the spectrum, margin and human involvement fall while automation and standardisation rise. The distribution model's job is to push standard flow rightward toward automation, where cost-to-serve approaches zero, and to reserve human coverage for the left end, where a relationship and judgement still win the business. Choosing the channel is not an operational detail; it is the economic decision at the heart of the model.

Electronification has advanced unevenly, and liquidity sets the pace

Not every market electronified at the same speed, because standardisation and liquidity decide what a machine can price. Foreign exchange is the most electronic market of all — spot and forwards trade overwhelmingly on streaming prices and algorithms, sub-second. Cash equities are close behind, order-book-driven, with algorithms and direct market access dominant and sales-trading focused on large or complex blocks. Rates are high and rising, with government bonds, futures, and vanilla swaps trading heavily on request-for-quote and streaming platforms. Credit is electronifying fastest from a low base, transformed by portfolio trading and all-to-all. Bespoke OTC derivatives and structured products remain largely human. The shares below are illustrative orders of magnitude that shift constantly, but the ranking is stable and tells the story.

Figure 4 Approximate electronic share of trading by asset class — illustrative orders of magnitude, not precise figures. Liquidity and standardisation set the pace: FX and cash equities lead, rates follow, credit is catching up fast, structured products lag.

In Europe the pace was pushed by policy, not only by markets. MiFID II deliberately drove flow onto regulated venues — a trading obligation for shares and certain derivatives, and a regime for systematic internalisers — so electronification in Europe is partly a designed outcome. That matters for anyone assessing a European distribution business: the electronic share of a market is not just a technology trend the bank is riding, it is a regulatory direction the bank is required to follow, and being connected to and compliant on the right venues is a standing obligation rather than a competitive choice.

Step 3 — The client asks for a price, and the request itself has become a protocol

How a trade gets done electronically depends on the liquidity of the instrument

Once a client wants to trade, the mechanism it uses is called an execution protocol, and each protocol suits a different liquidity profile. Request-for-quote is the workhorse of bond and swap e-trading: the client asks one or several dealers for a price on a specific trade, the dealers respond, and the client picks the best. Streaming, or click-to-trade, is dominant in FX and liquid rates: dealers continuously stream live prices and the client clicks to deal instantly, with no negotiation. A central limit order book matches anonymous buyers and sellers by price and time priority, exchange-style, for the most liquid instruments. Portfolio trading lets a client buy or sell a whole basket of bonds in one transaction against a single dealer price — a credit-market game-changer. All-to-all lets any participant respond to any other, so the buy-side can provide liquidity too, expanding the pool. Auction and matching sessions concentrate liquidity at a point in time to move large or illiquid interest with less market impact.

Protocol How it works Best for Effect on the sales dynamic
RFQ Ask one or more dealers, pick the best price Bonds, swaps — the e-trading workhorse Dealer still sees the client and can add value
Streaming / click-to-trade Continuous live prices, click to deal FX, liquid rates Price-only, near-anonymous
CLOB Anonymous order book, price/time priority The most liquid instruments Fully anonymous, pure price
Portfolio trading A whole basket in one trade, one price Credit — moving many line items at once Relationship plus pricing engineering
All-to-all Any participant quotes any other Less-liquid credit, expanding the pool Buy-side becomes a competitor liquidity source
Auction / matching Periodic sessions concentrate liquidity Large or illiquid interest Reduces market impact for size

Step 4 — For liquid flow, a machine now makes the price

The pricing engine has replaced the human quote at the fast end

On standard, liquid flow the bank no longer quotes by hand; a pricing engine does it, thousands of times a second. The engine takes in live market data and reference prices, the desk's current axe and inventory, the client's tier and relationship, the desk's risk limits and current position, and its own win-rate data against the competition. From those inputs it computes a fair mid, applies a spread calibrated to the client and the size, skews the price toward the desk's axe, checks the risk and the limits, and decides in milliseconds whether to quote, to pass, or to auto-hedge. The output is either a live streamed price or an instant RFQ response, an automatic execution and booking, an optional auto-hedge, and a record of the win or loss that feeds back to sharpen the next quote.

Figure 5 The auto-pricing engine: market data, the desk's axe, the client tier, risk limits, and win-rate data converge into a computed quote in milliseconds — the human quote replaced for liquid flow. Illustrative.

Building and running that engine is hard for four reasons, and each is a permanent engineering discipline. Latency, because a slow price loses the trade — milliseconds decide who wins. Calibration, because a price too wide never wins and a price too tight loses money, so the spread has to be exactly right for the client and the size. Data quality, because bad market data produces bad or dangerous quotes, and the engine will act on whatever it is fed. And risk control, because auto-quoting can accumulate a position frighteningly fast and needs hard, automatic limits that a human could never enforce at machine speed. The pricing engine is often the crown jewel of a bank's e-distribution — usually built in-house or heavily customised, precisely because it encodes the desk's own edge.

The pricing engine is where the front-office technology stack earns its keep, and it never stands alone. It consumes market-data feeds (expensive, high-volume, latency-sensitive, and licence-controlled), reads the desk's real-time risk and inventory, connects out to every venue over FIX, and writes every quote and fill to the systems that book, risk-manage, and report the trade. A weakness anywhere in that chain — a slow feed, a stale limit, a dropped session — shows up as a mispriced quote or a missed trade. This is why "distribution as an engine" is as much an infrastructure problem as a modelling one.

Inside the price: mid, spread, and skew

A quote is not one number but three decisions stacked on top of each other, and separating them is the key to understanding electronic pricing. The first is the mid — the engine's best estimate of fair value right now, built from live market data, related instruments, and reference prices. The second is the spread — how far from the mid the bank quotes, a function of the instrument's liquidity, the size of the trade, the volatility of the moment, and the client's tier. The third is the skew — a deliberate tilt of the price to one side, driven by the desk's axe and current position. If the desk is long a bond and wants to sell, it skews the offer sharper and the bid wider, nudging flow toward the side that reduces its risk. A client sees a single price; the engine sees a fair value, a margin for risk and relationship, and a lean that reflects what the desk needs to do with its book.

Figure 6 Inside the pricing engine: five inputs feed a five-stage pipeline — compute the mid, apply the spread, skew for the axe, check risk and limits, decide to quote, pass, or hedge — and every stage is logged. Illustrative.

The three decisions carry different kinds of risk, which is why the engine treats them separately. A wrong mid is a data problem — feed the engine stale or bad market data and it will quote a fair value that is simply incorrect, and it will do so at machine speed across thousands of quotes. A wrong spread is an economics problem — too wide and the bank never wins, too tight and it wins unprofitable trades. A wrong skew is a risk problem — skew the wrong way and the engine accumulates exactly the position the desk was trying to shed. Tiering sits across all three: a top-tier client like Meridian is quoted a tighter spread than an unknown account, because the relationship is worth more than the marginal trade, and the engine encodes that judgement as a parameter the salesperson sets. Understanding a quote as mid-plus-spread-plus-skew is what separates someone who can manage a pricing engine from someone who merely reads its output.

What the machine does to the salesperson's job

When the engine prices, the human moves up the value chain rather than out of it. For vanilla flow the salesperson is out of the pricing loop entirely — the engine quotes and executes. What the salesperson now does is manage the engine: set the client tiers, the spreads, and the axe skews that the engine applies; handle the exceptions and the large trades the engine is not trusted to price; and cover the high-value relationships and complex business a machine cannot win. Distribution becomes part relationship business and part software product, and the salesperson becomes part relationship manager and part product manager for the pricing engine. This is the single most important shift in the modern sales floor, and it reframes what a "good salesperson" even means.

The competitive arena: where banks meet on the client's screen

Multi-dealer platforms turned distribution into an engineering problem

A multi-dealer platform lets a client request prices from several banks at once and trade with the best, all in one workflow. For the client the appeal is obvious: competition, efficiency, automatic capture, and best-execution evidence in a single place. For the bank the trade-off is sharp — reach and volume on one side, margin pressure and reduced relationship visibility on the other, because the interaction is price-led and the salesperson may never see the client at all. The dominant electronic channel for institutional FICC flow now runs through these venues: Tradeweb across rates, credit, ETFs, and derivatives; MarketAxess, the dominant corporate-bond platform that pioneered all-to-all and portfolio trading; Bloomberg's terminal-based RFQ, where enormous volumes originate; and FX-specific venues such as 360T, FXall, and EBS.

To win on a multi-dealer platform the bank must do three things well, and all three are technical. It must stream competitive prices automatically, respond to RFQs in milliseconds, and be connected to every relevant venue with the right certifications and rules mastered. That turns distribution into an engineering problem: pricing engines, low-latency connectivity, auto-quoting logic, and analytics to decide which RFQs are worth winning and at what price. A bank that is not connected to a venue simply cannot win the flow that runs on it — connectivity is not glamorous, but it is decisive.

Figure 7 An RFQ on a multi-dealer platform: the client fans one request to several banks, each auto-quotes in milliseconds, and the client trades on the best price. The bank's edge shifts from relationship to price and speed. Illustrative.

Under MiFID II many of these platforms are regulated venues — multilateral trading facilities or organised trading facilities — and a bank that internalises client flow against its own book falls under the systematic-internaliser regime, with its own pre- and post-trade transparency obligations. European venues such as Tradeweb Europe sit alongside exchange-linked infrastructure, and the post-Brexit split between EU and UK venues adds a layer of reporting and connectivity complexity that a cross-border bank must manage on both sides. The point for distribution is that "connect to the venue" is never purely a technology task; it is a regulated-venue onboarding with compliance obligations attached.

Single-dealer platforms are the bank's own storefront

A single-dealer platform is the bank's own portal, where it controls the experience and keeps the relationship. On it, clients see the bank's prices, axes, research, and analytics, and execute directly — everything branded and controlled by the bank. The strategic tension with multi-dealer platforms is real and unresolved. A single-dealer platform keeps the relationship, shows the full breadth of content, and avoids reducing the bank to one price in a list, so it carries higher margin and more loyalty. A multi-dealer platform wins on convenience, because clients like comparing many banks in one place. Most banks therefore do both: they stream to multi-dealer platforms for reach, and they invest in their own platform for high-value clients and differentiated content. A single-dealer platform only wins if it offers something the multi-dealer venues cannot — unique pricing, data, analytics, or service. Content, again, is the differentiator.

Market microstructure: spread, depth, and the danger of being right too often

The shape of liquidity decides how a trade moves the price

To price and distribute well, a desk has to understand market microstructure — the fine-grained mechanics of how liquidity sits in a market and how trades move prices. Three ideas carry most of it. The bid-offer spread is the immediate cost of a round trip, the gap between where you can buy and where you can sell right now. Depth is how much size sits near the touch — a market can have a tight spread but little depth, so a large order blows through it. Market impact is the price move a trade itself causes: buy in size and you push the price up against yourself, so the true cost of a large order is not the spread but the spread plus the impact of working it. These three together explain why a EUR 5 million trade and a EUR 500 million trade in the same bond are completely different problems, and why the second needs a sales-trader's judgement or a careful algorithm rather than a single click.

The subtlest and most important idea in microstructure is adverse selection, and it governs how a market-maker survives. When a bank streams a price and a client hits it, the bank has to ask an uncomfortable question: why did this client want to trade at exactly this level, right now? Sometimes the answer is innocent — the client had a portfolio need. But sometimes the client knows something the bank does not, and the trades a market-maker wins are disproportionately the ones informed counterparties wanted to do against it. That is adverse selection, and it is why a naive market-maker who quotes tight and wins everything goes broke: it is winning precisely the trades it should have lost. The defence is to price defensively against informed flow, to widen when the market is moving, and to read which clients carry toxic, information-rich flow versus benign, uninformed flow. This is the deep reason the win rate is not a measure of success — a desk that wins too much is being adversely selected, and the pricing engine's win-rate feedback exists precisely to detect it.

Axes and inventory: the information layer of distribution

An axe is a signal, and matching it to the right client is a data problem

An axe is a position the desk particularly wants to trade, and it is valuable information. If the desk holds a bond and wants to reduce that risk, it is "a better seller" of it, and it will usually offer a sharper price on that side. Showing an axe tells clients where the bank's best price is and helps the desk manage its inventory — so axes are simultaneously a service to the client and a risk-management tool for the desk. Related signals travel the same way: inventory, meaning what the bank holds and can offer; runs, the lists of prices a desk sends out; and indications of interest, signals of buying or selling interest, especially in equity blocks.

Broadcasting axes well is a genuine data-and-controls problem, not a mailing list. Axes and runs go out through chat, email, the single-dealer platform, and third-party aggregators, reaching clients where they already are. The prize is to match the right axe to the client most likely to want it — targeted, not spammed — which is an analytics job that reads client history and holdings. The prices and axes shown must be consistent across every channel and compliant, because a misleading signal is a conduct problem. And the feedback loop matters: what a client looked at, passed on, or traded is data that sharpens the next broadcast. Done well, axe distribution is a quiet engine of both revenue and inventory management; done badly, it is noise that clients learn to ignore.

The central risk book: internalising flow at scale

The most important recent evolution in electronic distribution

As electronic flow grew, banks discovered they could manage its risk far better by centralising it, and the result — the central risk book — is one of the most consequential recent developments in distribution. The problem it solves is simple to state. When hundreds of small electronic trades hit a bank across a day, each in principle needs hedging, and hedging each one individually into the market is expensive and leaks information. A central risk book aggregates the risk from all that flow into one place, where opposing trades net against each other automatically — a client buying from one desk offsets another client selling to another — so the bank only needs to hedge the residual, net position rather than every individual trade. The savings are large, and they come from two sources: less hedging cost, because most of the flow self-nets, and less market impact, because the bank is not constantly trading in the open market to lay off each ticket.

The central risk book turns a stream of trades into a single, continuously managed inventory, and that reframes what internalisation means. When a client buys and the bank can fill that buy from flow it already holds — because another client just sold — the trade never touches the external market at all; it is internalised. A bank with enough flow can internalise a large share of its business, capturing the spread on both sides and hedging only what does not net. This is why scale in electronic flow is self-reinforcing: more flow means more internal netting, which means lower costs, which means sharper prices, which wins more flow. The central risk book is where that flywheel lives, and it is a genuine fusion of a business idea — internalise and net — with a hard technology build — a real-time, firm-wide risk aggregation and hedging engine wired to every electronic channel.

The central risk book is only possible because of the data and integration architecture the rest of this module keeps returning to. To net a buy from one channel against a sell from another in real time, the bank needs every electronic trade flowing into one place the instant it happens — which is exactly what event streaming delivers and batch integration cannot. To hedge the residual intelligently, it needs a live, firm-wide view of its position, clean market data, and an automated hedging engine with hard risk limits. The central risk book is therefore a showcase of the whole stack working together: streaming integration, a real-time risk engine, clean data, and automated execution, all in service of a business idea that only pays off if the technology underneath it is genuinely real-time. It is the clearest example in distribution of technology architecture directly creating competitive advantage.

Equities are their own world: sales-trading and the algorithm

In cash equities the machine that trades is an execution algorithm

Cash equities electronified earliest and deepest, and the distribution model there looks different from the FICC picture. Equities trade on exchanges through a central limit order book, so the bank is usually an agent working the client's order into the market rather than a principal quoting a spread. The client can reach the market three ways: direct market access, in which the client's order routes straight to the venue under the bank's membership; sales-trading, in which a human works a large or sensitive order with judgement about timing and impact; and execution algorithms, in which software slices a big order into small child orders to minimise market impact. The algorithms have names that describe their logic — VWAP works the order to match the volume-weighted average price over a window, TWAP spreads it evenly over time, percentage-of-volume tracks a set share of market volume, and implementation-shortfall balances the cost of trading now against the risk of waiting. Choosing and tuning the right algorithm for an order is the modern equity execution skill.

Sales-trading is where the human still earns the commission, and it concentrates on what the algorithm cannot handle. A large block that would move the market if worked openly, an illiquid name with thin liquidity, or a client who wants a capital commitment — the bank taking the other side as principal to give the client immediate certainty — all need a human sales-trader's judgement. Much of this happens in venues designed to hide size: dark pools, where large orders match without pre-trade transparency so that the market does not see them coming. The equities distribution model is therefore a barbell: a vast, automated, low-touch algorithmic-and-DMA business for standard flow, and a high-touch sales-trading business for blocks and complexity, with the commission pool migrating steadily toward the automated end.

Equity execution is an OMS-and-EMS problem end to end, and it is one of the most demanding real-time systems in the bank. An order-management system holds the client order and its constraints; an execution-management system routes child orders across venues, chooses algorithms, and monitors fills in real time; a smart-order router decides, order by order, which venue offers the best execution at that microsecond. All of it depends on low-latency market data and exchange connectivity, and all of it must produce the time-stamped record that proves best execution afterward. When people say cash equities is a technology business with a sales team attached, this is the stack they mean.

Execution algorithms, in more detail

The execution algorithms that work large orders are worth understanding one level down, because the choice among them is a real distribution skill. Each embodies a different answer to the same trade-off: trade quickly and pay market impact, or trade slowly and bear the risk that the price moves against you while you wait. A volume-weighted-average-price algorithm works the order in proportion to expected market volume through the day, aiming to match the day's average price — good for a patient order in a liquid name. A time-weighted-average-price algorithm spreads the order evenly over a set window, ignoring volume — simple and predictable. A percentage-of-volume algorithm participates at a fixed share of whatever volume trades, speeding up when the market is active and slowing when it is quiet. An implementation-shortfall algorithm is the most sophisticated: it explicitly balances impact against timing risk, trading faster when the cost of waiting is high and slower when impact would dominate, to minimise total execution cost against the price when the decision was made. Choosing and tuning the right algorithm for a given order, client, and market is where the modern equity sales-trader adds value.

Foreign exchange: the fully electronic frontier

FX shows where every other market is heading

Foreign exchange is the most electronic market in the world, and studying it is the closest thing to seeing the future of distribution. Spot FX in particular is deep, standardised, and traded around the clock, so almost all of it is priced and executed by machine. Two streaming models dominate. Executable streaming prices are continuous, firm, clickable quotes that a client trades on instantly — the fastest, most automated form of distribution there is. Request-for-stream is a lighter negotiation for larger or less-liquid FX, where the client asks for a short-lived stream of prices to trade on. On top of this, banks and clients run execution algorithms for large FX orders, slicing them to minimise impact exactly as in equities. The whole apparatus runs on sub-second pricing, low-latency connectivity, and enormous automated volume, and the human salesperson is reserved for the large, the structured, and the relationship-defining.

FX also introduces a controversial mechanic worth understanding, because it shows the tension between speed and fairness — last look. Because FX streams are fired out to many clients at once and markets move in milliseconds, a bank quoting a firm streaming price faces the risk that a fast client trades on a price that has already moved. Last look is the practice of giving the bank a final, brief window to accept or reject a trade after the client has clicked, protecting it against being picked off on a stale quote. It is legitimate as a defence against latency arbitrage, but it can be abused if a bank uses the window to reject trades that have simply moved against it — and regulators and the FX Global Code have pushed hard for transparency about how and when last look is applied. For a distribution professional, last look is the perfect illustration of a recurring theme: as pricing goes fully automated, entirely new conduct questions appear that did not exist in the voice world, and the answer is always more transparency and tighter governance, not less automation.

Connectivity and market data: the plumbing behind every click

None of this works unless the bank is wired in, reliably and fast

Every electronic interaction rests on connectivity, and connectivity is unglamorous but decisive. A bank that is not connected to a venue cannot win the flow that runs on it, and a link that is slow or unreliable loses trades to faster competitors. Five pieces make up the plumbing, and each is a permanent workload. FIX sessions connect the bank to venues and clients as the standard language of orders and executions. Venue and platform connectivity means reliable, often low-latency links to every relevant multi-dealer platform, exchange, and electronic communication network — each with its own onboarding, certification, and rules. Client APIs let sophisticated accounts trade directly against the bank's prices through documented, secure, versioned interfaces. Market-data feeds bring real-time prices in from exchanges and vendors — the fuel for pricing, and expensive, high-volume, latency-sensitive, and licence-controlled. And monitoring with failover treats downtime as lost trades and reputational damage, making capacity and resilience first-class concerns rather than afterthoughts.

Market data deserves its own emphasis, because it is both the fuel and a major cost. The pricing engine is only as good as the data feeding it: stale or wrong market data produces stale or dangerous quotes, and the engine will act on whatever it is given at machine speed. Real-time feeds from exchanges and vendors are among the largest fixed costs of a markets business, licensed per user and per use, and policed for redistribution. So a distribution franchise runs a permanent balancing act — buy enough data, fast enough, to price competitively, without the data bill overwhelming the thinning margins the flow now earns. The economics of market data are, quietly, one of the constraints that shape which flows a bank can afford to price at all.

Connectivity is a large, never-finished engineering estate, and it is where a distribution technology programme spends much of its time. The FIX layer alone means maintaining hundreds of sessions, each with a counterparty's specific dialect and certification; the market-data layer means normalising many feeds into one clean internal picture; the API layer means supporting external clients who depend on the bank's uptime. Every venue added is a new integration to build and monitor, every client API a new support obligation, every data feed a new licence and a new latency budget. This is the permanent workload behind "the bank is connected" — and the reason connectivity, not modelling, is often where the modernisation money actually goes.

Marketing mechanics: how demand is tested before it is built

For a new issue, the book is de-risked before it opens

Before a bookbuild opens in public, the syndicate has usually already tested the water in private, and the mechanics have precise names. Pre-sounding — sometimes called pilot fishing — is the confidential canvassing of a handful of key investors before the deal is announced, to gauge appetite, price sensitivity, and structure. It carries an information-barrier consequence: an investor who is pre-sounded is brought "over the wall" and becomes an insider on the deal until it is announced, which is a controlled, logged event, not a casual call. Anchor and cornerstone orders are the next lever: a large, early, named commitment from a respected investor that gives the book its backbone and signals confidence to everyone who follows. Testing-the-waters, the broader version, lets the issuer and syndicate assess demand for an equity deal before formally launching. The common thread is that a good bookbuild is not a leap into the unknown; it is the public confirmation of demand the syndicate already believes is there.

Cross-border distribution: the same bond, two rulebooks

Where a deal is marketed decides which rules apply

A European distribution business rarely sells into one jurisdiction, and each border it crosses brings a different rulebook. The same Lumen bond may be placed with investors across the European Union, in the United Kingdom, and — through the appropriate exemptions — with US institutional accounts, and each of those touches a different regime. Within the EU, MiFID II and MiFIR govern conduct, transparency, and reporting, supervised by ESMA and national regulators. The post-Brexit UK runs a divergent version of the same rules, so a cross-border bank documents and reports the same trade two ways depending on where the client sits. Selling to US institutions without a full US registration relies on established private-placement exemptions, which impose their own eligibility and documentation requirements on who may be shown the deal. The distribution consequence is concrete: the client's location and classification decide which version of the offering it may see, how the trade is reported, and which venue it may trade on — so the "same" bond is, in compliance terms, several different products depending on who is buying it.

Step 5 and Step 6 — Executing, then placing: bookbuilding and allocation

For a new issue, execution is a bookbuild, and the book is a demand curve

Placing a new issue is the most concentrated form of distribution, and it runs on a bookbuild. When Lumen Energy brings its EUR 750 million seven-year bond, the syndicate announces the deal and sales calls and messages the target accounts with the credit story and initial price talk — a deliberately wide spread that opens the auction. Investors place orders, and as the book builds, sales feeds the demand and the price sensitivity back to syndicate in real time. The order book is a demand curve: it shows how much money wants the bond at each spread level. When the book is comfortably oversubscribed, the syndicate tightens the spread to the level that still clears with high-quality demand — the best outcome for the issuer. The quality of the book matters as much as its size: a book full of sticky, long-term real-money orders is worth more than a book of the same size stuffed with fast money that will sell on day one.

Allocation is where competing interests collide, and it must be documented

A well-covered book means someone gets less than they asked for, and deciding who is the delicate part. Allocation pits the issuer's interest in stable, long-term holders against the banks' interest in rewarding good clients against each investor's desire for a full fill. The criteria are consistent even if the judgement is not: reward early orders and price leadership, favour real-money accounts that will hold over fast money that will flip, protect key relationships, and keep the aftermarket in mind, because an allocation that leaves too much bad paper in weak hands sells off on day one. Crucially, allocation is no longer a matter of private judgement alone. Under MiFID II the process must be documented and justified — fair, non-preferential treatment that can be evidenced to a regulator — which turns allocation into a data-and-controls requirement and is exactly why electronic syndicate tools now capture orders and record the allocation rationale.

Inside the syndicate: roles, the pot, and stabilisation

A new issue is placed by a structured team with defined economics

Primary distribution is run by a syndicate, and the roles within it carry different responsibilities and different economics. The lead bookrunners — sometimes called global coordinators on the largest deals — run the process: they set the strategy, own the book, advise on price, and drive the allocation. Joint bookrunners share that work and the fee. Co-leads and co-managers play a lesser distribution role for a smaller share. The distinction is not only about prestige; it is about who controls the book and who therefore carries the reputational risk if the deal goes badly. On our running example, the bank is a joint bookrunner on Lumen's bond, which means it shares responsibility for building the book and gets a share of the fee proportional to its role and its distribution contribution.

How the fee is shared is governed by whether the deal runs on a pot or a retention basis, and the difference shapes behaviour. Under a pot system, orders go into a single common book that all bookrunners see, and the issuer can reward the banks that brought the best demand — it is transparent and pushes banks to place with genuine, high-quality investors. Under a retention system, each bank is allocated a fixed slice to place with its own clients, which is simpler but weaker at rewarding real distribution skill. Most institutional benchmark deals run on a pot, precisely because transparency of demand is what lets the issuer price tightly and allocate fairly. The mechanics of the fee are therefore also a mechanism for surfacing honest information about demand — a recurring pattern in which the structure of an incentive is really a structure for revealing truth.

After a deal prices, the bank's job is not over, because a new issue often needs stabilisation and aftermarket support. In equities, the mechanism is formal: an over-allotment option, the greenshoe, lets the syndicate sell extra shares and then buy them back if the price falls, cushioning the aftermarket for a defined window. In bonds there is no greenshoe, but the secondary desk supports the new issue by making markets in it, so that the investors who bought in primary can trade out — the aftermarket liquidity that justified their participation in the first place. Either way, the principle is the same: distribution does not end at pricing, because the promise a bank makes to a primary investor is a promise about the secondary market that follows, and honouring it is what earns the next mandate.

The syndicate technology stack: placing a deal is now a data exercise

Primary distribution has quietly electronified too

Secondary trading gets the attention when people talk about electronification, but primary distribution has been quietly transformed by technology, and the change matters as much for conduct as for efficiency. A modern bookbuild runs on a syndicate platform that captures every order electronically as it comes in — the account, the size, the limit, the timestamp — rather than in the scribbled pads and spreadsheets of the past. That order-capture system feeds a live view of the book that the syndicate and sales can watch build in real time, and it feeds the allocation tool that records not just who got what but why. Around it sit deal-management systems that coordinate the documentation, the timetable, and the many banks in a syndicate, and investor-data systems that tell the syndicate which accounts to target with which credit. Placing Lumen's bond is, underneath the phone calls, a data exercise: capture the demand cleanly, watch the curve build, price against it, and allocate defensibly.

The reason this is a conduct story as much as an efficiency one goes back to allocation fairness. When MiFID II required that allocation be fair, documented, and justifiable to a regulator, it turned the syndicate's order book from a private working document into a compliance record. The electronic syndicate platform is what makes that possible at scale: it captures the order flow as immutable, time-stamped data, records the allocation rationale against each account, and produces the evidence trail that shows long-term holders were rewarded and flippers scaled back for consistent, defensible reasons. So the same technology that makes bookbuilding faster also makes it auditable — and a bank running its primary business on spreadsheets is not just inefficient, it is exposed, because it cannot evidence the fairness the rules now demand. Primary distribution is the clearest case of a theme that runs through the whole module: the technology that improves the business and the technology that satisfies the regulator are, increasingly, the same technology.

The primary stack shares the data problem of the secondary side, seen from a different angle. An order in a bookbuild carries the same core fields as any other client interaction — the account, resolved to a legal entity and classification; the size; the timestamp — and it feeds the same downstream needs: allocation, settlement, reporting, and the client-history record that sharpens the next deal's targeting. A bank whose primary and secondary systems do not share a client identity ends up with two partial views of the same account — its new-issue behaviour in one system, its secondary flow in another — and cannot see, for instance, that an account which anchors every new deal also trades actively in secondary and therefore deserves priority. The golden client record of Part 4 is what joins the two, which is why unifying primary and secondary client data is one of the quieter but higher-value prizes in a distribution transformation.

Best execution and TCA: proving the client got a fair deal

A regulatory duty that became a data discipline

When a bank executes a client order it must take all sufficient steps to get the best possible result, and it must be able to prove it. Under MiFID II, best execution weighs price, cost, speed, likelihood of execution, size, and settlement — it is not simply "best price." For a large or illiquid order, certainty of execution and low market impact can matter more than the headline level. Firms must have an execution policy, monitor their outcomes, and demonstrate to clients and regulators that they achieved best execution. That obligation reshaped distribution: it made the capture of clean, time-stamped trade and market data a first-order requirement rather than an operational nicety.

Transaction cost analysis is the measurement discipline that answers the duty. It grades execution quality against benchmarks such as arrival price or volume-weighted average price, quantifying the cost and quality of each trade. Both sides use it: the buy-side to grade its brokers, the sell-side to prove and improve its own execution. It is data-heavy, needing clean trade, timestamp, and market-data capture across venues — a serious data-engineering task. And it feeds the relationship: transparent, favourable execution statistics become a sales argument, evidence that the bank serves the client well. Best execution is where a compliance obligation turns into a commercial edge, provided the data underneath it is clean.

Step 7 — The trade settles, and the relationship rolls on

Settlement must not fail, and after it the service begins

A trade is not real until it settles, and for the flow that runs on platforms, settlement should be untouched by human hands. The goal is straight-through processing: an electronic trade flows automatically from execution into risk, booking, confirmation, clearing, settlement, and reporting, with no manual re-keying anywhere along the chain. Every re-key is a chance for an error and a cost to carry, so the entire industry pushes toward end-to-end automation. For standard secondary flow this is largely achieved; for new issues and complex trades, pockets of manual handling remain, and closing them is a permanent programme.

Between execution and settlement sits the post-trade chain, and a distribution professional should know its shape even though another team runs it. Once a trade is done, it is confirmed — both sides agree the economics — and then it clears and settles. For exchange-traded and many standardised products, a central counterparty steps into the middle of the trade, becoming buyer to every seller and seller to every buyer, so that neither side bears the other's default risk; this is novation, and it is why a central counterparty concentrates and manages counterparty risk for the whole market. Settlement itself moves the security against the cash on a delivery-versus-payment basis — neither leg completes unless both do — through central securities depositories and the custodians that hold assets for investors. The settlement cycle runs to a standard timetable, T+1 or T+2 depending on the market, and a trade that fails to settle on time incurs cost and, increasingly, a regulatory penalty. The distribution relevance is simple: the client experiences a failed or delayed settlement as the bank's failure, so the clean hand-off from front-office execution into this post-trade chain is part of the service, not a separate department's problem.

After settlement, step 7 is where the relationship is renewed. The client gets reporting and best-execution evidence, after-sales support on anything that needs it, and — the part that matters most commercially — the next idea. Distribution is not a series of discrete trades; it is a continuous relationship in which each settled trade is the occasion for the next conversation. The salesperson who services well, reports cleanly, and comes back with a relevant idea keeps the wallet share that the whole model competes for. The arc closes where it began, at content and idea, and the loop starts again.

  1. Name the seven steps of the sales workflow and identify where effort and risk peak.

  2. What is the organising principle behind the channel spectrum from voice to API?

  3. How does the choice of execution protocol change where a dealer's edge lives?

  4. What five inputs feed a pricing engine, and why is auto-quoting hard?

  5. In cash equities, what are the three ways a client reaches the market, and what does a sales-trader still do?

  6. Why is a new-issue order book best understood as a demand curve, and why does its quality matter as much as its size?

  7. Why is allocation now a data-and-controls requirement, not just a judgement call?

  8. What does best execution require beyond "best price," and how does TCA answer it?

  9. What role does a central counterparty play in post-trade, and what does delivery-versus-payment guarantee?

  10. What is a central risk book, and why does scale in electronic flow become self-reinforcing?

2

  1. Where do effort and risk peak in the sales workflow?

  2. What principle organises the channel spectrum from voice to API?

  3. Flow moves from RFQ to anonymous streaming. Where does the dealer's edge move?

  4. Which input does NOT feed a pricing engine?

3Roles, Actors, and Interactions

Distribution is a team sport played across a wall. This part names every actor — the investors who supply the demand, the front-office teams that face them, the controls that govern who may do what, and the venues and vendors in between — and shows why the salesperson sits in the middle of nearly every interaction. It pays special attention to two things a generalist usually underestimates: how different the investors are from one another, and how much of the operating model exists to keep information from flowing where it must not.

Four camps surround every interaction, and the bank sits in the middle

Strip distribution down and you find four groups plus one referee

Every distribution interaction involves four camps, and the bank front office touches all of them. The first camp is the buy-side client — asset managers, hedge funds, insurers, pension funds, banks, corporates — the demand the whole model serves. The second is the bank's own front office: sales and sales-trading own the relationship and distribute prices, ideas, and axes; the trading desk prices risk and provides liquidity; syndicate runs primary distribution; and e-commerce runs the platforms, pricing engines, and connectivity. The third camp is controls: compliance polices suitability, best execution, allocation fairness, and market abuse, while client onboarding and KYC form the gate a client must pass before it can trade at all. The fourth is the market infrastructure between the bank and the client — trading venues, exchanges, data providers, and the CRM and OMS/EMS vendors whose systems the flow runs on. Over all of it sits the referee: the regulator, present in every interaction through the rules it imposes even when it is not in the room.

The mental model is an orchestration around a single client relationship. The front office faces the client, controls govern who and how, and venues and vendors sit between the bank and the buy-side. Distribution is the act of coordinating all three around one relationship at once — which is why the coverage salesperson, who owns that relationship, is the natural conductor, and why the operating challenge the model solves is keeping a whole team pointed at the same client.

Figure 8 The distribution ecosystem: the bank's front office at the centre, with the buy-side, the desk and syndicate, the controls functions, and the venues and vendors around it — and the regulator over all. The salesperson is party to nearly every interaction. Illustrative.

Who talks to whom, and when

The interactions follow a stable pattern across the workflow. The table below maps the major information and instruction flows — who originates each, who receives it, and at which step it happens. Read it as the wiring diagram behind the seven-step flow of Part 2.

From To What flows Stage
Research / strategy Sales Views, colour, trade ideas (across the wall) Content & idea
Desk Sales Axes, inventory, live prices Content & pricing
Sales Client Ideas, axes, quotes, runs Reach-out & pricing
Client Sales / platform Enquiry, RFQ, order Enquiry
Sales Desk / pricing engine Request for a price Pricing
Syndicate Sales Deal announcement, price talk, allocation Primary distribution
Client Onboarding / KYC Legal, classification, documentation Before any trade
Desk / e-commerce Venue Streamed prices, RFQ responses, executions Execution
Front office Compliance Comms, trades, allocation records Continuous surveillance
Ops CSD / custodian Settlement instructions, confirmations Settlement

The market infrastructure, in depth

Between the bank and the buy-side sits a layer of regulated plumbing

The fourth camp — market infrastructure and vendors — is more structured than "venues and vendors" suggests, and a distribution professional should know its parts. On the trading side sit the venues: regulated exchanges for listed instruments, and, under the European rules, multilateral trading facilities and organised trading facilities for other flow, alongside banks acting as systematic internalisers when they trade client flow against their own book. On the transparency side sit the reporting utilities: approved publication arrangements that publish trades to the market, and approved reporting mechanisms that submit transaction reports to regulators. On the post-trade side sit the risk and settlement utilities: central counterparties that novate and net trades to manage default risk, central securities depositories that hold and settle securities, and custodians that safekeep assets for investors. Each is a regulated entity with its own onboarding, connectivity, and rules — which is why "connect to the market" is never a single task but a portfolio of integrations, each with a compliance dimension.

The vendor layer sits alongside this regulated plumbing, and it splits into recognisable categories the reader has now met. There are the data-and-terminal hubs, dominated by Bloomberg, that carry market data, chat, and RFQ flow. There are the multi-dealer platforms — Tradeweb, MarketAxess, and the FX venues — where banks compete. There are the system vendors for CRM, OMS, EMS, and surveillance. And there is a growing band of specialist fintechs and RegTech providers filling narrow, fast-moving niches from TCA to comms surveillance to KYC. The strategic point is that a distribution business does not stand alone; it is a node in an ecosystem of regulated infrastructure and commercial vendors, and its competitiveness depends as much on being cleanly connected to the right parts of that ecosystem as on anything it builds itself. Mapping this ecosystem — who does what, which parts are regulated, where the bank connects — is the first thing a newcomer to a distribution technology programme has to learn.

The buy-side is not one investor but several, and the differences drive everything

Why the type of investor matters as much as the size of the order

A salesperson tailors everything — the idea, the channel, the price, the allocation — to the type of client. The buy-side looks homogeneous from the outside, but its members have opposite horizons, opposite sensitivities, and opposite value to the bank, and treating them alike is the fastest way to lose them. The main types, and how each behaves, are worth learning precisely, because they recur in every later decision — especially allocation, where knowing the investor is the other half of the decision.

Asset managers are the real-money backbone of a book

Long-only asset managers — "real money" — are large, benchmark-driven, and buy-and-hold. They run funds against an index or a mandate and represent the ultimate, stable demand in most markets. They place the big tickets and the anchor orders in new issues, and because they hold rather than flip, issuers prize them and banks reward them in allocation. They are the core of most distribution, and a EUR benchmark bond lives or dies on real-money demand. Meridian, our running example, is one of these.

Hedge funds bring speed, sophistication, and fees

Hedge funds are active, leveraged, fast-trading, and often short as well as long. They demand speed, ideas, and financing, they trade in high value and high frequency, and they are among the biggest fee payers in the business. They also carry flipping risk in new issues — the fast money that can sell into the aftermarket — so a salesperson values their flow and their fees while watching their behaviour in a book. Kestrel, our other running example, is one of these: sophisticated, price-sensitive, and living on electronic execution.

Pension funds and insurers are the patient giants

Pension funds and insurers have very long-term liabilities to match, and that shapes their demand. They are huge, rate- and inflation-sensitive, and natural buyers of long bonds, swaps, and structured solutions — the patient capital that anchors long-dated issuance. In Europe they are especially important: the continent has vast insurance and pension pools, and specific rules such as Solvency II shape how much and what kind of risk insurers can hold, which in turn shapes their demand. A salesperson covering them thinks in decades, not days.

Banks, central banks, and the official sector are large and rate-driven

Bank treasury and asset-liability desks, central banks, and reserve managers are large, rate-driven counterparts. They manage liquidity and reserves rather than seek alpha, they are important buyers of high-quality paper, and they are significant liquidity counterparts in rates and FX. The official sector in particular moves in size and with signalling weight, and being trusted to handle its flow discreetly is its own franchise.

Private banks and wealth are a distinct, suitability-heavy channel

Private banks and wealth managers distribute onward to wealthy individuals, and they are big buyers of structured products and funds. They are a distinct channel with a distinct sensitivity: because the end client is often a retail or near-retail investor, suitability and disclosure rules bite hard. In Europe this is a large market — Switzerland, France, Luxembourg — and it is governed by the PRIIPs key-information-document regime for packaged retail products. A salesperson covering this channel sells through an intermediary and must respect the protections the end investor is owed.

Corporates and retail sit at the edges of the flow

Corporates hedge through the derivatives desk, and retail reaches markets through brokers and platforms. Corporates are not investors in the portfolio sense; they use the markets business to hedge rates, FX, and commodity exposures, and they are covered by a corporate-facing sales team. Retail investors access markets indirectly, through brokers, online platforms, and structured notes, and they bring breadth, complexity, and a political dimension — the fair treatment of retail is a matter regulators watch closely.

Investor type Holding horizon Price sensitivity Why the bank wants or fears them
Asset managers (real money) Long, buy-and-hold Moderate Backbone demand and anchor orders; prized in allocation
Hedge funds Short, fast-trading High High fees and flow; flipping risk in new issues
Pension funds & insurers Very long Moderate, yield-driven Patient anchors for long-dated issuance
Banks & central banks Varies, rate-driven High Large liquidity counterparts; discreet size
Private banks & wealth Medium, product-driven Lower, suitability-bound Big structured-product demand; heavy suitability duty
Corporates & retail Hedging / indirect Varies Hedging flow; retail breadth and political scrutiny
The universe is wider than six types, and the edges matter

Beyond the core types, several fast-growing investor groups have their own behaviour, and a modern distributor ignores them at its peril. Passive and ETF managers now command enormous assets and buy mechanically, tracking an index rather than forming a view — their demand is large, price-insensitive within their mandate, and driven by fund flows, so understanding the ETF creation-and-redemption mechanism is now part of distributing both equities and, increasingly, bonds. Private credit and direct-lending funds have grown into a major force, lending directly to companies outside the public markets and competing with the bank's own financing business even as they consume its distribution and financing services. Family offices and ultra-high-net-worth vehicles behave like small, idiosyncratic institutions with real size and long horizons. And sovereign wealth funds are patient giants that move in scale and prize discretion. The common lesson is that the buy-side is not a fixed list but an evolving ecosystem, and the accounts growing fastest — passive, private credit — are precisely the ones reshaping where distribution revenue will come from next.

Figure 9 The buy-side placed on two axes that drive coverage: holding horizon and price sensitivity. Patient, price-insensitive investors sit top-left; fast, price-sensitive ones sit bottom-right — and each quadrant is covered, priced, and allocated differently. Illustrative.

The coverage model turns the client list into a plan

Tiering, account plans, and the discipline of scarce attention

Coverage is not a matter of answering the phone; it is a deliberate allocation of scarce human attention across a client base, and it runs on a few disciplines. The first is tiering: clients are sorted by their value — wallet, profitability, strategic importance — into tiers that determine how they are served, from intensive senior coverage for the top tier down to a mostly electronic, self-service relationship for the long tail. The second is the account plan: for each important client, the coverage salesperson maintains a living picture of the relationship — what the client trades, where the bank wins and loses, what its people care about, what the whitespace is — and a plan to grow the bank's share of its wallet. The third is the call report: the record of every meaningful interaction, feeding the client history that the analytics and the next-best-action models depend on. These are unglamorous disciplines, but they are how a franchise turns a list of names into a managed portfolio of relationships.

The reason this matters technologically is that every one of these disciplines is only as good as the data beneath it, which is why coverage and the data platform are the same project. Tiering requires a clean, unified view of what each client is actually worth across all products — the very number that fragmented data makes impossible to compute. Account planning requires the whole relationship in one place, not a credit-desk view and a rates-desk view that never meet. Call reports require capture that salespeople will actually use, which means it has to be effortless. When a bank says it is investing in its client-data platform, what it is really buying is the ability to run its coverage model as a disciplined portfolio rather than a set of disconnected personal relationships — and that is the difference between a franchise that compounds and one that merely churns.

Inside a buy-side client, several people place the trade

Knowing who you are talking to is half the coverage skill

A buy-side client is not a single decision-maker, and a salesperson who treats it as one will misread it. In a typical asset manager the portfolio manager decides what to buy or sell, driven by the fund's mandate and view; the buy-side dealer, or execution trader, actually places the order and chooses how and with whom to execute it; and a research or strategy function feeds the portfolio manager ideas. The salesperson deals mostly with the buy-side dealer on execution but must reach the portfolio manager with ideas, because that is where the decision — and the new-issue demand — originates. On a new bond, the portfolio manager decides to participate and the buy-side dealer works the allocation and any secondary trading around it. Reading which conversation belongs to which person, and carrying the right content to each, is a large part of what makes coverage skilled rather than merely friendly.

The internal split on the buy-side mirrors, in miniature, the split on the sell-side, and that symmetry matters. Just as the sell-side separates the person who owns the price (the trader) from the person who owns the relationship (the salesperson), the buy-side separates the person who owns the decision (the portfolio manager) from the person who owns the execution (the dealer). Best execution lives on the buy-side dealer's side of that line — it is the dealer who must justify the choice of counterparty and venue — which is why the sell-side's execution quality, TCA, and platform presence are aimed as much at the buy-side dealer as at the portfolio manager. Sell to the decision, execute with the dealer, and keep both sides of the client's house satisfied: that is the shape of a well-covered relationship.

Coverage and specialist sales pull in opposite directions, by design

The tension between the relationship and the product is the operating challenge

The two axes of the sales floor — coverage by client and specialist by product — are in permanent, productive tension, and managing it is the core operating challenge. The coverage salesperson owns the whole relationship and thinks client-first: what does Meridian need across everything, and how do we grow our share of its wallet? The specialist thinks product-first: how do I distribute the credit desk's axes and win the most credit flow, across all my clients? Both are right, and they can collide — the coverage view might hold back a hard credit sell to protect a broader relationship, while the specialist wants to push it. The bank resolves this not by choosing one axis but by making a team of both: a coverage salesperson orchestrating the relevant specialists around each client, with the trading desk, sales-trading, and syndicate alongside. Getting that team to act as one, around one client, with one current picture of the relationship, is precisely the problem the operating model — and the unified client data of Part 4 — exists to solve.

Information barriers: the wall that runs through the whole institution

One institution knows things it must not share with itself

A bank is a single legal entity that routinely holds information some of its own people must never see. A deal team advising Lumen on its bond knows material non-public information about Lumen; the sales floor distributing bonds must not. Research forms independent views that sales distributes but cannot direct. The private side works on confidential mandates; the public side trades and talks to clients all day. Left unmanaged, these adjacencies would let inside information leak from the people who legitimately hold it to the people who could trade or talk on it — which is exactly what insider-dealing and market-abuse rules forbid. The information barrier, historically called the Chinese wall, is the control that separates these functions inside one institution.

Four functions in particular are held apart. Research is walled from sales and trading so that its views stay independent and are not shaped by the bank's positions. The private side — deal teams holding confidential client information — is walled from the public side that faces the market. Sales and trading are separated from the compliance and control functions that police them. And where a person legitimately needs to cross the wall — bringing a public-side salesperson "over the wall" to help sound out investors on a confidential deal — that crossing is a formal, controlled, logged exception, not a casual conversation. Wall-crossing is the managed doorway through an otherwise solid wall.

The information barrier is a technology problem before it is a policy one, because at scale a policy no human can enforce is only enforced by systems. The stack that holds the wall up includes: restricted and watch lists that flag names the bank cannot trade or research; access controls that partition systems and data so that private-side information is technically unreachable from the public side; communication surveillance that reads chats, emails, and voice for leakage across the wall; wall-crossing logs that record every controlled exception with a timestamp and a reason; and personal-account-dealing systems that monitor employees' own trades against the restricted list. None of this works on trust alone — it works on entitlements, logs, and monitoring, which is why the wall is a data-and-controls architecture, not a memo.

Six internal teams run a distribution franchise

Each team owns a piece of the flow, and the wall runs through the table

A distribution business is staffed by six recognisable teams, and knowing each one's division, role, and technology makes the org chart legible. The salesperson — coverage or specialist — owns client relationships, distributes ideas, prices, and axes, and executes and services; it is the revenue-facing core. The sales-trader or trader prices and executes, manages the book and the liquidity, and, in the sales-trader's case, bridges sales and the desk, especially in equities. Syndicate runs primary distribution, building books, setting price talk, and allocating new issues across accounts. E-commerce, usually sitting under a chief operating officer, owns the electronic channels, the pricing engines, the platform connectivity, and the automation strategy — an increasingly central function. Compliance and controls police suitability, best execution, allocation fairness, communications surveillance, and market abuse, a gate on everything. And IT, data, and change owns the distribution stack — CRM, pricing, connectivity, platforms, analytics, surveillance — the place where transformation lands.

Team Division Core role Primary technology
Coverage / specialist sales Front office Own relationships; distribute ideas, prices, axes; execute CRM, chat, single-dealer platform
Sales-trading / trading Front office Price and execute; manage book and liquidity Pricing engine, OMS/EMS, risk systems
Syndicate Front office (primary) Build books, set price talk, allocate new issues Electronic bookbuilding tools
E-commerce / e-trading (COO) Front office / COO Run platforms, pricing engines, connectivity, automation Auto-pricing, FIX, venue connectivity
Compliance & controls Control Suitability, best-ex, allocation fairness, surveillance Surveillance, restricted lists, reporting
IT / data / change Technology Own and modernise the distribution stack CRM, data platform, analytics, connectivity

Read the table against the wall and against the workflow and two things stand out. First, the information barrier runs straight through it: research does not appear as a front-office team here because it is walled off, feeding content across the barrier rather than sitting inside distribution. Compliance sits over the whole table as a control, not beside it. Second, the rise of e-commerce has made the technology axis — the COO and IT teams — as important as the sales desk itself. A modern distribution franchise blends people, a trading book, heavy technology, and strict conduct rules, and the balance has tilted toward technology. Modernising the stack — CRM, e-pricing, data, surveillance — is the recurring work, which is why the operating model and the systems architecture of Part 4 are really the same subject seen from two angles.

  1. Name the four camps in a distribution interaction and explain why the salesperson sits in the middle.

  2. Contrast a real-money asset manager with a hedge fund on horizon, price sensitivity, and value to the bank.

  3. Why is investor type the missing half of the allocation decision?

  4. What problem does the information barrier solve, and why is it a technology problem rather than only a policy one?

  5. Which internal team has become as central as the sales desk, and why?

  6. What are the three disciplines of the coverage model, and why is each a data problem?

3

  1. Why does the coverage salesperson sit in the middle of a distribution interaction?

  2. Real-money manager versus hedge fund: which statement is right?

  3. Why is investor type the missing half of the allocation decision?

  4. Why is the information barrier a technology problem, not only a policy one?

4Systems, Data, and IT

Underneath the relationships and the platforms runs a stack of systems, and the health of that stack decides whether the model works. This part opens it up in the right order: first the clean question of what functions any distribution business needs, then who builds or buys each one, then the data that flows through them, the identifiers that tie it together, the protocols that carry it, and the resilience that keeps it alive on the day it must not fail. The recurring theme is that almost every weakness in distribution traces back to data that does not join up.

Start with functions, not vendors

Any distribution business needs the same five functional blocks

Before naming a single product, ask what functions a client interaction must pass through, and the architecture answers itself. Whatever the bank, a distributed trade flows through five blocks in sequence. Client data and CRM hold who the client is — history, wallet, interactions — the 360-degree view that targets coverage. Pricing and content produce what sales takes to the client: pricing engines, axes and inventory, research distribution. Channels and order-and-execution management carry the interaction and route the trade: voice, chat, platforms, APIs, and the OMS/EMS that manages orders and executions. Execution and booking do the trade, on-venue or internalised, and capture and risk-manage it. Controls and reporting close the loop: surveillance, best-execution and TCA, transaction reporting, and client and regulatory reporting. Client and market data feed the whole chain, and a client, a price, and a trade must stay consistent from one end to the other.

Functional block What it does Consequence if it fails
CRM & client data Holds the 360-degree client view; targets coverage Reactive, uncoordinated coverage; missed opportunities
Pricing & content Produces prices, axes, and research to distribute Uncompetitive or inconsistent prices
Channels & OMS/EMS Carry the interaction and route the order Lost flow; un-captured communications
Execution & booking Executes, captures, and risk-manages the trade Mis-booked risk; broken straight-through processing
Controls & reporting Surveillance, best-ex, transaction and client reporting Regulatory breach; unprovable best execution
Figure 10 The distribution functional architecture: five blocks from client data to controls, over a cross-cutting band of client and market data, connectivity, and compliance that every block depends on. Illustrative.

Underneath the five blocks runs a band of cross-cutting capabilities that every block depends on. Client and market data feed all five. Connectivity — to venues, platforms, and clients — carries the flow in and out. Compliance and the information barrier govern the whole chain rather than sitting in one box. Draw the architecture this way, as blocks over a shared band, and the failure modes become obvious: they cluster where the blocks meet and where the cross-cutting band is thin. Fragmented client data, channel silos, inconsistent pricing across channels, and a heavy reporting load pulling from many systems are all seam problems, not block problems — which is exactly why unifying the seams is where transformation creates value.

Who builds and who buys: the vendor and platform stack

The buy-versus-build line is a map of where the edge lives

Every functional block is filled by some mix of vendor product and in-house build, and where a bank builds tells you where it thinks its edge is. The commodity functions are bought; the differentiating ones are built. The table below names the systems and venues a distribution professional meets daily and what each is really for. Read the final column as the strategic tell: a bank buys the terminal and the venues because everyone has them, and builds or heavily customises the pricing engine and the client-data platform because those are where it competes.

System / venue Category What it is for Buy or build
Bloomberg Hub Terminal, market data, IB chat, huge RFQ flow Buy — hard to distribute without it
Tradeweb / MarketAxess Multi-dealer platforms Institutional RFQ, portfolio trading, all-to-all Buy / connect — core electronic venues
CRM (vendor or proprietary) Client data Client relationships, interactions, coverage Mix — a major unification target
OMS / EMS Order/execution management Route orders, manage execution, feed STP Mix — vendor or in-house
Pricing & e-trading engines Pricing Stream prices and answer RFQs Build — the crown jewel of e-distribution
Surveillance & reporting Controls Comms capture, trade surveillance, TCA, reporting Mix — increasingly specialist vendors

The pattern is consistent across banks and worth stating as a principle. Where a function is a shared utility — market data, a trading venue, comms capture — the bank buys it, because building a worse copy of a standard would waste money and connectivity. Where a function encodes the desk's own judgement — how to price, how to skew for an axe, how to decide which RFQ to win — the bank builds it, because that judgement is the product. And where a function is a growing regulatory burden — surveillance, transaction reporting — the market has produced specialist vendors, and banks increasingly buy rather than carry the cost of keeping pace with the rules. The buy-versus-build map is therefore not static; functions migrate from build to buy as they commoditise, and the crown jewels are guarded.

Order and execution management, and the smart router

Between the client and the venue sits a layer of routing intelligence

When a client order needs to reach a market, it passes through an order-management and execution-management layer, and understanding the division of labour clarifies a lot. An order-management system holds the order and its constraints — the client, the instrument, the size, the limits, the compliance checks — and is the system of record for what the client asked for. An execution-management system is the fast, market-facing tool that actually works the order into the venues: choosing an algorithm, slicing the order, monitoring fills in real time, and adjusting as the market moves. In many equities businesses the two are distinct systems from distinct vendors; in FICC they are often more integrated. The order lives in the OMS; the execution happens in the EMS; and the two must stay in sync so that what was executed matches what was ordered, down to the last child fill.

The intelligence that decides where each slice of an order goes is the smart order router, and it is where best execution becomes concrete. In a fragmented market where the same instrument trades on several venues, the router decides, order by order and often microsecond by microsecond, which venue offers the best combination of price, liquidity, and cost right now — and splits the order accordingly. It is the software embodiment of the best-execution duty: not a policy document but a live routing decision, logged so that the bank can later prove it took all sufficient steps. When Part 2 said best execution made clean, time-stamped data a duty, the smart order router is where that duty is discharged in real time, and its logs are the evidence. A distribution business that cannot show why each order went where it did has a best-execution problem hiding in its routing layer.

Data lineage: follow one interaction from CRM to the regulator

An error upstream reaches every system downstream

Trace a single client trade through the stack and the cascade principle becomes concrete. A client interaction starts in CRM, where the client is identified and its history and classification live. The pricing block reads the client's tier and the desk's axe and produces a quote. The channel carries the RFQ and the response; the client deals. Execution captures the trade — instrument, price, size, timestamp, venue — and books it into risk. From that captured trade, the controls block generates the transaction report to the regulator, the best-execution and TCA record, and the surveillance data. Every downstream object depends on an upstream one: the transaction report is only as correct as the trade capture, which is only as correct as the client identity that CRM supplied. An error at the top — a mis-identified client, a wrong classification, a stale reference price — does not stay at the top; it flows all the way to the regulatory report and the surveillance alert.

Figure 11 Data lineage across distribution: client identity and interaction in CRM feed pricing, which feeds the executed and booked trade, which feeds transaction reporting, best-execution records, and surveillance. An error upstream cascades to every downstream system. Illustrative.
Stage System Data object created What depends on it
Interaction CRM Client record, classification, tier The price, the suitability check, the report
Pricing Pricing engine Quote (mid + spread + skew) The executed level, best-ex evidence
Execution OMS/EMS, venue Trade (price, size, time, venue) Booking, reporting, surveillance
Booking Risk / booking system Position, P&L, risk Limits, hedging, valuation
Reporting Controls stack Transaction report, TCA, surveillance record The regulator's view of the trade

The client-data platform: from many records to one

Entity resolution is the technical heart of the 360-degree view

The most valuable and most difficult system a distribution business builds is the one that turns many scattered client records into a single, trustworthy view — and it deserves to be understood as architecture, not aspiration. The raw material is fragments: the credit desk's view of the client, the rates desk's view, the FX platform's logs, the chat archive, the research-readership data, the revenue and profitability numbers — each held in a system built for its own purpose and none sharing an identifier. The platform's job is to ingest all of it, resolve which fragments belong to the same client through entity resolution, and produce a golden client record that every downstream system can trust as the single source of truth. On top of that golden record sit the things the business actually wants: the 360-degree view, next-best-action, an accurate wallet-share number, and cross-channel surveillance.

Figure 12 The client-data platform: fragmented records from every desk, channel, and system are resolved into one golden client record, which then feeds the 360-degree view, next-best-action, wallet share, and surveillance. Resolving one client from many records is the atom of the whole model. Illustrative.

Entity resolution is genuinely hard, and understanding why explains why the 360-degree view has been promised for two decades and delivered rarely. A single client faces the bank through a hierarchy of legal entities, funds, and sub-accounts, whose names are spelled inconsistently, whose relationships change as the client reorganises, and whose identifiers — where they exist at all — were assigned independently by systems that never spoke. Resolving Meridian into one identity means deciding, often probabilistically, that a dozen differently-named records across a dozen systems are the same economic client, and then keeping that resolution current as the client and the data both change. This is master-data management, and it is a data-engineering problem of the first order — which is exactly why "unify the client data" is easy to say, expensive to do, and the foundation on which every analytics and AI ambition in the business ultimately rests. A bank without a golden client record is not one model away from data-driven distribution; it is one hard, unglamorous data platform away from it.

The atom of distribution is the client-and-order record

The core data object is not the trade alone; it is the join between the client and the order, and its fields drive everything downstream. A client record carries the legal entity and its hierarchy, the MiFID classification (retail, professional, or eligible counterparty), the coverage team, the tier and wallet, and the agreements in place. An order or RFQ record carries the instrument identifier, the side and size, the client, the channel, the timestamp, and the price. Each field drives something: classification drives which products may be shown and which protections apply; tier drives the spread the engine applies; timestamp drives best-execution and surveillance; the instrument identifier drives settlement and reporting. Get a field wrong and the error propagates — a mis-classified client can be shown a product it should never see, and the breach surfaces only later in the reporting.

Identifiers are the joins that make the data resolvable

Three identifiers do most of the work of tying the data together, and each bites at a different point. The Legal Entity Identifier names the counterparty — the who — and is mandatory for transaction reporting; without a valid LEI, a client cannot be reported and, in practice, cannot trade. The International Securities Identification Number names the instrument — the what — and drives settlement, reporting, and reference data. The Market Identifier Code names the venue — the where — and is required in transaction reporting to say where a trade happened. Alongside them, the client classification is not an identifier but a status flag that travels with the client and governs what may be distributed to it. Get the identifiers right and the data joins cleanly across systems; get them wrong and every downstream join fails.

Identifier Names Where it bites
LEI (Legal Entity Identifier) The counterparty Transaction reporting; no LEI, no trade
ISIN (Int'l Securities Identification Number) The instrument Settlement, reporting, reference data
MIC (Market Identifier Code) The venue Transaction reporting — where the trade happened
MiFID classification The client's regulatory status What may be distributed to whom, and which protections apply

Integration and messaging: how the systems actually talk

Batch, API, and streaming are three ages of integration

Systems connect to one another in three patterns, and a bank usually runs all three at once because it modernised in layers. Batch is the legacy default: data moves in scheduled bulk transfers, simple and robust but latency-bound, so downstream systems see yesterday's or this-morning's picture and reconciliation is needed to catch what drifted. APIs are the modernisation: a system asks another for exactly what it needs, on demand, in real time. Event streaming is the frontier: systems publish events — a quote, a fill, a limit change — that interested systems consume the instant they happen, Kafka-style, which reduces the need for reconciliation because everyone sees the same event at the same time.

The migration from batch to streaming is more than a speed upgrade, and this is the point most people miss. Batch integration forces reconciliation: because two systems updated at different times from different copies of the data, a nightly process must check they still agree, and a team must chase the breaks. Streaming removes the need for that reconciliation, because both systems consumed the same event and never diverged in the first place. So moving a distribution stack from batch to streaming does not just make it faster; it removes an entire category of operational work and the errors that come with it. That is why "real-time architecture" is a cost-and-risk story, not a latency vanity project — and why it is worth the expense of rebuilding plumbing that already technically works.

Two messaging worlds: FIX in front, SWIFT and ISO 20022 behind

The front office and the back office speak different languages, and the split is the front-to-back divide made concrete. In the front office, FIX — the Financial Information eXchange protocol — is the industry standard for orders and executions. The bank connects to venues and clients over FIX sessions, and it is the lingua franca of electronic trading: fast, real-time, and built to carry an order to a venue and a fill back in milliseconds. In the post-trade world, the older SWIFT MT message types carry settlement instructions and confirmations between banks, custodians, and central securities depositories, and that estate is migrating to the richer, more structured ISO 20022 (MX) standard. The two worlds meet at the seam where a front-office execution becomes a post-trade settlement instruction.

Layer Standard Carries Timescale
Front office FIX Orders, quotes, executions Milliseconds, real-time
Post-trade SWIFT MT → ISO 20022 (MX) Settlement instructions, confirmations Seconds to T+1/T+2

The pricing engine is becoming a product with an API

Open architecture turns distribution into a platform business

The most advanced distribution franchises no longer think of their pricing as an internal function; they expose it as a product through APIs. When a bank publishes a documented, secure, versioned interface to its prices, a sophisticated client can wire the bank directly into its own trading systems and trade against those prices automatically, with no human and no screen in between. That changes the relationship into something closer to a software integration: the bank competes on the quality, latency, and reliability of its API the way a technology company competes on its developer experience. Once a client has built to a bank's API, switching is costly, so a good API is also a retention tool — the modern, technical version of a sticky relationship. The same openness runs inward: banks increasingly build their own stacks from composable services, so that the pricing engine, the risk check, and the booking step are separate components wired together, rather than one monolith, which lets them replace or upgrade a piece without rebuilding everything.

This is why distribution is described, only half in jest, as a platform business with a sales team attached. The bank's prices, content, and execution are increasingly delivered as services — over an API, through a single-dealer platform, into a multi-dealer venue — and the engineering that delivers them reliably is as much the product as the prices themselves. It also reframes the make-versus-buy decision from Part 4's vendor discussion: a bank that has built clean internal APIs can adopt a new vendor component quickly, while one whose systems are tangled cannot, so architecture quality becomes a competitive variable in its own right. The direction is clear and one-way: the more distribution electronifies, the more it becomes a business of well-designed, well-run software delivered to clients as a service, with human coverage concentrated where that service still needs a person behind it.

FIX in depth: sessions, messages, and why it won the front office

FIX became the lingua franca of electronic trading because it standardised two things at once — how systems stay connected, and what they say to each other. It works in two layers. The session layer keeps a reliable, ordered, authenticated connection alive between two parties, handling logon, heartbeats, sequence numbering, and the recovery of missed messages, so that neither side ever silently loses a message. The application layer carries the business content as tagged fields — a message is a series of tag-equals-value pairs, where each tag is a number with an agreed meaning. The workhorse messages are few and worth knowing by name: a New Order Single sends an order, an Execution Report reports a fill or a status change, a Quote streams a price, and an Order Cancel Request pulls an order back. A client trading against the bank over FIX is, concretely, exchanging these messages over a session that must never drop.

Figure 13 The message lifecycle of a trade: born in the FIX world of orders and executions (milliseconds), it grows up in the SWIFT and ISO 20022 world of settlement instructions and confirmations (seconds to T+2). Straight-through processing lives or dies at the FIX-to-ISO seam. Illustrative.

The practical weight of FIX is that every venue and client speaks a slightly different dialect of it, and reconciling those dialects is a permanent workload. The standard is common, but each counterparty uses its own subset of tags, its own custom fields, and its own certification process, so onboarding a new venue or client over FIX is a small engineering project, and a bank maintains hundreds of such sessions at once. When the connectivity section of Part 2 called this a large, never-finished estate, the FIX layer is much of what it meant: not the protocol itself, which is stable, but the sprawl of slightly-incompatible implementations that a distribution business has to keep certified, monitored, and alive.

SWIFT and ISO 20022 in depth: the post-trade language and its migration

Once a trade leaves the FIX world, it enters the messaging world of the back office, and that world speaks SWIFT. The legacy SWIFT MT standard carries the post-trade instructions between banks, custodians, and central securities depositories as structured message types with numeric names: the securities-settlement family — instructions to receive or deliver securities against payment — and confirmation messages that acknowledge what was agreed. These messages are older, more rigid, and less data-rich than modern needs demand, which is why the whole industry is migrating to ISO 20022, whose MX messages carry the same instructions in a richer, more structured, more extensible XML format. The migration is a multi-year, industry-wide programme, and for a distribution business it matters because the richer data of ISO 20022 reduces the ambiguity and manual repair that plague the older format.

The reason to care about a back-office standard from a front-office seat is that the two worlds must hand off cleanly, and the handoff is where automation succeeds or fails. A trade executed over FIX in milliseconds must become a settlement instruction in the SWIFT or ISO world within the settlement cycle, and if that translation needs a human to fix it, straight-through processing has broken and cost and risk have entered. The migration to ISO 20022 is, in part, an effort to make that handoff cleaner by carrying more structured data end to end, so that fewer trades fall out for manual repair. This is why a distribution technologist tracks a settlement-messaging migration that seems, on the surface, to belong to someone else's department: it is the far end of the same straight-through chain that begins at the pricing engine.

Resilience: the day the system must not fail

Some moments in distribution have no second chance

Distribution has moments that cannot be rerun, and the systems must survive them. A pricing engine that goes down during a volatile session is not quoting, which means it is not winning trades and may be leaving the desk with un-hedged risk it cannot see. A connectivity failure to a venue on a busy morning silently hands the flow to competitors. A new-issue pricing that has to happen at a set hour cannot wait for a system to come back up. These are the irreversible moments, and for them ordinary "it usually works" reliability is not enough — the architecture must be built to keep running through a failure, not merely to recover afterward.

Concept What it guarantees Why it matters on the critical day
Redundancy No single point of failure A dead server does not stop the quoting
Automatic failover Instant switch to a standby The engine keeps pricing through a fault, unattended
Disaster recovery A tested recovery to a second site A site loss does not end the trading day
Business continuity People and process keep running The desk operates when the primary site is unreachable

Market-data architecture: the pipeline that feeds every price

Clean, fast data is the fuel, and moving it is an engineering discipline

Every price the bank makes rests on market data, and getting that data from the exchange to the pricing engine fast and clean is a discipline in its own right. The pipeline has recognisable stages. Raw feeds arrive from exchanges and vendors in dozens of proprietary formats. Feed handlers decode each one into something the bank's systems can read. A normalisation layer maps everything into a single internal format and enriches it with reference data, so that the same instrument looks the same wherever it appears. A low-latency distribution bus then carries the normalised data to the consumers that need it — the pricing engine above all, but also risk, TCA, and surveillance. Each stage adds latency, and in a business where a slow price loses the trade, every microsecond in that pipeline is contested.

Figure 14 The market-data pipeline: raw exchange and vendor feeds are decoded by feed handlers, normalised into one internal format, and distributed over a low-latency bus to the pricing engine, risk, and TCA. Clean, fast data is the fuel — and one of the largest fixed costs of the business. Illustrative.

Two forces make market data one of the hardest and most expensive parts of the stack — latency and licensing. Latency drives banks to physically co-locate their pricing engines next to the exchange's matching engine, because the speed of light through fibre is a real constraint when microseconds decide who wins a quote; the whole architecture of a fast trading business is shaped by the geography of where the data is. Licensing makes the data a major fixed cost: exchanges and vendors charge per user and per use and police redistribution tightly, so a bank runs a permanent balancing act between buying enough data, fast enough, to price competitively and letting the data bill overwhelm the thin margins the flow now earns. When people say electronic distribution is a technology business, the market-data pipeline — expensive, latency-bound, and unglamorous — is a large part of what they mean, and it quietly constrains which flows a bank can afford to price at all.

The gate before any of this runs: client onboarding

A client cannot trade until it clears a heavy, slow process

Before a client sends its first RFQ, it must pass an onboarding gauntlet, and that gate is a major friction and cost point. Six steps stand between a new relationship and a first trade. KYC and anti-money-laundering checks verify the client's identity, ownership, and legitimacy and screen for sanctions and financial-crime risk. MiFID classification sorts the client into retail, professional, or eligible counterparty, which determines its protections and what may be sold to it. Legal and documentation puts the agreements in place — terms of business, ISDA and CSA for derivatives, platform agreements — that permit trading. Suitability and appropriateness assessments decide whether products fit the client, especially complex or structured ones. Credit and risk setup establishes the lines, limits, and collateral terms so the desk can face the client. And system and connectivity setup creates the client in every system and enables the channels and APIs it will use.

Onboarding is slow, manual, and costly — often weeks — and a poor experience loses business before it starts. It is also the point where several of the identifiers and data objects from earlier in this part are first created: the LEI is validated, the classification is set, the client identity that everything downstream depends on is established. Get onboarding wrong and the data quality problem is baked in from the first day. This is why digitising and accelerating onboarding — data reuse, workflow automation, digital KYC — is one of the highest-value, most common transformation programmes in the client-facing business, and why it recurs in Part 5.

Modern architecture: event streaming and the move to cloud

The stack is being rebuilt around events and elastic compute

The distribution stack is undergoing an architectural shift as consequential as electronification itself, and it has two linked strands — event streaming and cloud. The first strand replaces batch integration with event streaming. Instead of systems exchanging bulk files on a schedule and reconciling the differences afterward, they publish events — a quote, a fill, a limit change, a client update — onto a shared stream that any interested system consumes the instant it happens. Because every system consumes the same event, they never diverge, and the reconciliation that batch integration requires simply disappears. This is not a speed upgrade dressed up; it removes an entire category of operational work and the errors that come with it, which is why a bank will spend real money rebuilding plumbing that already technically functions.

Figure 15 Batch versus streaming integration. Under batch, systems update from separate copies at different times and must be reconciled, with a team chasing the breaks. Under event streaming, every system consumes the same event and never diverges, so no reconciliation is needed. Illustrative.

The second strand moves the stack to the cloud, but selectively, because latency and elasticity pull in opposite directions. The latency-critical parts of distribution — the pricing engine, the venue connectivity, the market-data handlers — stay physically close to the exchanges, on-premises or in co-located data centres, because the speed of light will not be argued with. But the elastic parts — client-data platforms, analytics, next-best-action models, surveillance, regulatory reporting, and increasingly generative AI — move to the cloud, where compute can scale up for a heavy overnight analytics run and down again, without the bank owning the peak capacity year-round. The result is a hybrid target architecture: a fast, fixed core near the market, wired by event streams and APIs to an elastic, scalable analytics-and-data estate in the cloud.

Figure 16 The hybrid target architecture: latency-critical trading — pricing, connectivity, execution — stays close to the market, while elastic analytics, client data, surveillance, and GenAI move to the cloud, connected by an event bus and APIs. Illustrative.

The cloud migration reframes the buy-versus-build map from earlier in this part. A bank whose systems are already decomposed into clean, API-connected services can adopt a new cloud analytics tool, or swap a vendor component, in weeks; a bank whose systems are a tangle of point-to-point batch feeds cannot, whatever its budget. So architecture quality — how modular and event-driven the estate is — becomes a competitive variable in its own right, because it sets how fast the bank can absorb the next innovation. This is the deep reason "modernise the platform" keeps appearing in distribution strategy: it is not the modernisation itself that pays, but the optionality it buys to adopt everything that comes after, cheaply and quickly.

Reference data and surveillance: the quiet backbone and the constant watch

Reference data is the backbone nobody notices until it breaks

Underneath every price and every trade sits reference data — the slow-moving facts about instruments, clients, and venues that everything else assumes are correct. An instrument master holds each security's identifiers, terms, and characteristics; a client master holds each counterparty's legal entity, hierarchy, and classification; a venue and calendar master holds where and when things trade. None of this is glamorous, and precisely because it is assumed rather than examined, it is a common root cause of downstream failure. A wrong maturity date on an instrument, a stale classification on a client, a missing identifier — each is a small error in a reference record that surfaces as a mispriced quote, a rejected report, or a failed settlement much later and far away. Mastering reference data — one authoritative, clean, current copy that every system reads — is one of the least visible and most valuable things a distribution technology function does.

Surveillance is the system that watches everything, and it is getting smarter

A distribution business is required to watch itself continuously, and surveillance is the system that does it. Two streams feed it. Trade surveillance monitors executions and orders for patterns of market abuse — insider dealing, front-running, manipulation such as spoofing or layering. Communications surveillance captures and monitors the calls, chats, and emails that must be recorded under the conduct rules, watching for collusion, information-barrier breaches, and misconduct. The hard part is that abuse rarely shows up in one stream alone: it appears as a suspicious trade explained by a suspicious message, so surveillance must correlate communications and trades together, across channels, at enormous volume — millions of messages and trades a day.

The classic failure of surveillance is drowning in false positives, and the response is a shift toward machine learning. Naive rules — flag every large trade, every mention of a keyword — generate so many alerts that analysts cannot review them, and real risk hides in the noise. Smarter, increasingly model-based surveillance narrows the alerts to those that actually matter, learning the difference between normal and anomalous behaviour rather than applying a blunt threshold. Senior-manager accountability regimes raise the stakes further, making named individuals responsible for conduct and therefore for the surveillance that evidences it. Surveillance is thus both a large data-engineering problem — capture, storage, correlation, search — and a growing analytics problem, and it is one of the fastest-growing technology domains in the whole client-facing business.

The mechanics of how surveillance actually reads communications are worth a closer look, because they are moving from crude to genuinely intelligent. The first generation was lexicon-based: flag any message containing words from a list — the name of a restricted stock, phrases like "keep this quiet". It caught the obvious and drowned analysts in false positives, because language is ambiguous and context-free keyword matching cannot tell a joke from a confession. The second generation applies natural-language processing to read intent and context rather than words alone, scoring a message for genuine risk and clustering related messages into a story. The frontier is holistic surveillance, which fuses communications and trading data into a single model: an alert fires not because a chat mentioned a stock, but because a suspicious message is explained by a suspicious trade that followed it. Holistic surveillance is hard precisely because it requires the unified data the rest of this part keeps returning to — you cannot correlate comms and trades that live in systems which never joined up.

Five regulatory regimes shape every system in the chain

Read each rule as a system specification, not a legal summary

Five regimes touch nearly every system in a distribution stack, and the useful way to read them is as requirements the architecture must meet. MiFID II and MiFIR are the master framework: they specify client classification, suitability and product governance, best execution and its evidence, pre- and post-trade transparency, the venue and systematic-internaliser regimes, transaction reporting, and communications capture. The Market Abuse Regulation (MAR) specifies surveillance: the bank must monitor for insider dealing, front-running, and manipulation across both trades and communications, which mandates cross-channel capture and correlation. PRIIPs specifies disclosure for packaged retail products — a mandatory key information document — which the distribution systems for structured products must generate and deliver. GDPR specifies how client personal data may be held and used, bounding the CRM and analytics stack and colliding, at times, with the record-keeping the other rules demand. And the transaction-reporting and best-execution technical standards specify the exact fields, identifiers, and timeliness of what must be reported — the reason LEIs, MICs, and clean timestamps are not optional.

Regime What it specifies System consequence
MiFID II / MiFIR Classification, suitability, best-ex, transparency, reporting, comms capture Touches every block; the master specification
MAR (Market Abuse Regulation) Surveillance of trades and communications Cross-channel capture and correlation engine
PRIIPs Key information document for packaged retail products Automated KID generation and delivery
GDPR Lawful holding and use of personal data Bounds and governs the CRM and analytics stack
Transaction-reporting / best-ex standards Exact fields, identifiers, timeliness Mandatory LEI, MIC, and clean timestamps

In Europe these regimes are supervised by ESMA and national regulators — in France, the AMF and the ACPR — and the post-Brexit UK runs a divergent version. A cross-border European bank therefore builds its systems to satisfy two regimes on the same desks, which is a data-and-configuration burden as much as a legal one. The consolidated point is that compliance is not a layer bolted onto distribution; it is a set of specifications designed into every system in the chain, from the classification field in CRM to the timestamp on the transaction report.

Four more regimes complete the picture

Beyond the five master regimes, four more shape distribution systems directly, and a serious IT view of the business has to include them. EMIR — the European Market Infrastructure Regulation — governs derivatives: it mandates central clearing of standardised derivatives, imposes margin on the rest, and requires both sides of every derivative trade to be reported to a trade repository, which is a large, exacting data obligation on any desk that distributes swaps. SFTR — the Securities Financing Transactions Regulation — imposes similar reporting on repo and securities-lending activity, the financing business that sits alongside distribution. CSDR — the Central Securities Depositories Regulation — introduced settlement discipline, including penalties for trades that fail to settle on time, which turned settlement efficiency from an operational nicety into a measured, penalised metric. And DORA — the Digital Operational Resilience Act — is the newest and, for a technologist, the most pointed: it makes operational resilience a binding regulatory requirement, mandating that firms manage their technology risk, test their resilience, and oversee their critical third-party providers, turning the resilience disciplines described earlier in this part into a matter of law.

Figure 17 Which regime touches which system block. Compliance is not a layer bolted on; each regime is a set of specifications designed into specific systems — from classification in CRM to reporting under MiFIR, EMIR, and DORA's resilience mandate. Illustrative.

Read together, these regimes make a single point that a distribution technologist should internalise: regulation is a system specification, and it is expanding. Each regime translates into concrete requirements on data, identifiers, timeliness, testing, and record-keeping — the LEI that EMIR reporting demands, the timestamps that best execution needs, the settlement discipline CSDR penalises, the resilience testing DORA mandates. And the direction is one-way: every few years another regime arrives, and it always resolves into more data captured, more precisely, and retained longer. This is why the controls-and-reporting block of the architecture is one of the fastest-growing parts of the stack, and why banks increasingly buy specialist RegTech rather than carry the cost of keeping pace in-house. Compliance is not the opposite of the business; in a distribution franchise it is a permanent, growing line of the engineering budget.

  1. Name the five functional blocks of a distribution stack and explain why failures cluster at the seams.

  2. What does the buy-versus-build line reveal about a bank's strategy?

  3. State the cascade principle and give one example of an upstream error reaching the regulator.

  4. Why is deduplication of client identity the hardest data problem, and what does it block?

  5. Why is moving from batch to streaming integration more than a speed upgrade?

  6. Distinguish the FIX world from the SWIFT/ISO 20022 world, and say where they meet.

  7. Why is redundancy alone insufficient for resilience on the critical day?

  8. Distinguish the FIX session layer from the application layer, and name two workhorse FIX messages.

  9. Why is moving latency-critical trading to the cloud the wrong call, and what does move?

  10. What made operational resilience a legal requirement, and what does entity resolution unlock?

4

  1. Where do failures cluster in a distribution stack?

  2. What do banks usually build rather than buy in distribution?

  3. A client is mis-classified in the CRM. What can follow?

  4. Why is deduplicating client identity the hardest data problem?

5Risks, Pain Points, and Transformation

A business that touches every client and every trade fails in characteristic ways, and it is worth knowing them before the transformation pitch. This part lays out the six risks a distribution franchise runs, the handful of metrics that separate a good desk from a lucky one, and three real scandals that each expose a different failure mode. Then it turns to where the model is heading — electronification and data-and-AI — and applies one honest filter to every trend: does this change the economics or the risk, or does it just change the appearance?

The risks of distribution are six, and they map to controllability

A MECE taxonomy of what goes wrong

Distribution runs six distinct risks, and separating them cleanly is the first step to controlling them. Market and inventory risk is the risk that the desk's warehoused positions move against it — the price of providing liquidity. Execution and best-execution risk is the risk that a client order is handled poorly or cannot be shown to have been handled well, breaching the duty of best execution. Conduct and mis-selling risk is the risk that a product is sold to a client for whom it is unsuitable, or that a client is treated unfairly. Reputational and relationship risk is the risk that the bank's conduct — or the company it keeps in a deal — damages the trust the whole franchise rests on. Conflict-of-interest risk is the risk that the bank favours itself or one client over another, most visibly in allocation and in the axes it shows. And operational and settlement risk is the risk that a trade fails to capture, book, or settle cleanly. The table names each, who owns it, and its main defence.

Risk type What it is Who owns it Main defence
Market / inventory Warehoused positions move against the desk Trading desk Risk limits, hedging, auto-limits on the engine
Execution / best-ex Order handled poorly or unprovable Sales-trading, compliance Execution policy, TCA, clean data capture
Conduct / mis-selling Unsuitable product or unfair treatment Sales, compliance Suitability checks, product governance, KID
Reputational / relationship Conduct or bad company damages trust Senior management Due diligence, culture, escalation
Conflict of interest Bank favours itself or one client Compliance, syndicate Allocation policy, documentation, the wall
Operational / settlement Trade fails to capture, book, or settle Operations, IT Straight-through processing, reconciliation, resilience
Figure 18 The six distribution risks arranged by controllability: operational and execution risks yield to engineering, while conflict and reputational risks live in judgement and culture. Illustrative.

Telling a good desk from a lucky one

Five metrics that measure distribution quality

Volume flatters a distribution business, so the metrics that matter measure quality and profitability, not size. Wallet share is the anchor: the bank's slice of each client's total spend, and the truest measure of relationship strength. RFQ win rate — the hit ratio — measures how often the bank wins the trades it quotes on, but it must be read carefully. Cost-to-serve measures what it costs to cover a client, and set against revenue it yields client profitability, the number that decides where coverage and balance sheet go. Best-execution quality, measured by TCA, grades how well client orders were actually executed. And for primary distribution, aftermarket performance — how a new issue trades in the days after pricing — measures whether the deal was placed in the right hands. The table sets out each with its rule of thumb.

KPI What it measures Rule of thumb
Wallet share The bank's slice of a client's total spend Top-three in a key client's wallet is the goal
RFQ win rate (hit ratio) Share of quoted trades won Too high can mean pricing too tight
Cost-to-serve / client profitability Revenue minus the cost of coverage Some clients cost more than they pay
Best-execution quality (TCA) Execution vs benchmark Consistent, not occasional, outperformance
Aftermarket performance How a new issue trades post-pricing Stable-to-firm signals a well-placed deal

Three real failures, three distinct modes

The cases are not random; they are the three ways distribution breaks

Distribution fails in three ways — through conduct, through the company it keeps, and through the market — and three real episodes show each. They are chosen because each maps to a different step of the process and teaches a different lesson, and together they cover the failure space.

The FX benchmark scandal — a failure of conduct and conflict

Between roughly 2013 and 2015, traders at several of the world's largest banks were found to have colluded to manipulate foreign-exchange benchmark rates, and the penalties ran to more than ten billion dollars across regulators. Traders in different banks coordinated in private chat rooms — with names that later became infamous — to move the benchmark rates against which client orders were fixed, and to front-run those orders. In November 2014 the US CFTC alone imposed over 1.4 billion dollars across five banks; in May 2015 five banks pleaded guilty to US criminal charges and agreed to pay more than 5.7 billion dollars; UK and other regulators added billions more. The failure was not a market accident. It was a conduct-and-conflict failure at the heart of distribution: the people entrusted to execute client orders used the information those orders gave them to enrich the bank and themselves. The lesson is that distribution is a trust business, and the same information that makes a salesperson valuable is the information that, misused, destroys the franchise — which is exactly why communications surveillance and the information barrier exist.

The 1MDB bonds — a failure of the company you keep

Between 2012 and 2013 a major bank arranged and distributed 6.5 billion dollars of bonds for the Malaysian state fund 1MDB, much of which was embezzled, and in 2020 the bank paid more than 2.9 billion dollars to resolve the resulting investigations. The bonds were raised in three deals, sold to investors, and the proceeds were looted with the complicity of bankers inside the arranging firm. The distribution failure here is distinct from the FX case: the mechanics of placing the bonds were not the problem — the problem was placing them at all, for a client and a purpose that proper due diligence should have stopped. The 2.9 billion dollar settlement spanned the US DOJ and SEC, the Federal Reserve, UK, Singapore, Hong Kong, and Malaysian authorities. The lesson maps to the very first step of the workflow and to the onboarding gate: distribution puts the bank's name and reputation behind everything it places, so who the bank agrees to serve, and what it agrees to sell, is a risk decision before it is a revenue one. KYC and reputational due diligence are not friction to be minimised; they are the control that stops a fee from becoming a catastrophe.

The March 2020 dash for cash — a failure of the market itself

In March 2020, as the pandemic hit, the demand to hold cash overwhelmed even the most liquid markets, and for a period distribution simply seized. Investors rushed to sell assets for cash, including normally ultra-liquid government bonds; dealers, constrained by their own balance sheets and risk limits, could not absorb the flow; bid-offer spreads blew out and, in stretches, two-way markets vanished. The corporate-bond market effectively froze. Distribution failed not because anyone misbehaved but because the liquidity the whole model promises — the assurance that an investor can always trade out — evaporated at once. It took extraordinary central-bank intervention to restore functioning. The lesson maps to the aftermarket-support and market-making role: the liquidity a bank provides is real but finite, and it is scarcest exactly when it is most needed. A distribution franchise that has quietly assumed liquidity is always available has mispriced its most important promise.

Case Failure mode Maps to Lesson
FX benchmark scandal (2013–15) Conduct and conflict of interest Execution; the information barrier Trust is the asset; misused client information destroys the franchise
1MDB bonds (2012–13; settled 2020) The company you keep Onboarding; content and idea Who and what you agree to distribute is a risk decision first
March 2020 dash for cash The market itself Aftermarket support; secondary liquidity Liquidity is finite and scarcest when most needed

The pain points are mostly one problem wearing six masks

Where distribution actually hurts

Ask any distribution technologist where the business hurts and the same six problems recur, and five of them are the same problem underneath. Fragmented client data means trades, chats, RFQs, research reads, and CRM notes live in silos, so there is no single 360-degree view and coverage stays reactive. Margin and commission pressure, from electronification and MiFID II, squeezes the economics and forces relentless efficiency. Channel and pricing inconsistency means the price and axe shown by voice, on the single-dealer platform, and on a multi-dealer platform can diverge, and keeping them consistent is hard. Connectivity and venue sprawl means maintaining reliable, compliant links to many venues and clients is a large, never-finished job. Compliance and surveillance load means best-execution, transaction reporting, comms capture, and market-abuse monitoring demand huge, ever-changing data pipelines. And slow, manual onboarding means KYC, classification, and documentation take weeks, friction that loses business before it starts.

Step back and the pattern is unmistakable: most of these are data and integration problems. Unifying the client view, keeping prices consistent across channels, connecting to venues, and feeding compliance from clean data are all versions of the same underlying task — making data join up across systems that were never designed to talk. Only margin pressure is a genuinely external force; the other five are internal, and they are solvable with the right architecture. That is precisely why distribution transformation creates value under real economic pressure: the pain is concentrated exactly where good engineering helps most, and the bank that solves the data-and-integration problem solves five of its six pain points at once.

Where the model is heading

Transformation, trend by trend

The distribution model is being rebuilt around platforms, automation, and data, and five trends carry most of the change. Auto-pricing and auto-execution take the human out of the pricing loop for standard flow — the norm in FX, spreading across rates and credit — leaving sales to manage the engine and the exceptions. The growth of electronic protocols, especially portfolio trading and all-to-all, keeps expanding automation into less-liquid markets and reshaping who provides liquidity. Single-dealer platforms and apps let banks defend the relationship and deliver content and self-service beyond what multi-dealer venues offer. Digital onboarding and self-service cut the weeks-long friction of starting and maintaining a relationship. And straight-through processing removes manual re-keying end to end, taking out operational risk and cost. Each attaches to a specific step of the workflow, which is how to tell a real trend from a slogan.

Trend Attaches to Why it matters
Auto-pricing & auto-execution Price & execute Removes cost from standard flow; frees humans for the complex
Growth of electronic protocols Enquiry & execution Expands automation into less-liquid markets; changes who provides liquidity
Single-dealer platforms & apps Reach-out & service Defends the relationship and the margin against price-only venues
Digital onboarding & self-service The onboarding gate Cuts weeks of friction that loses business
Straight-through processing Execute & settle Removes operational risk and cost across the flow

Apply one filter to every transformation claim: does it change the economics or the risk, or does it only change the appearance? Auto-pricing changes economics — it removes cost from the flow and lets the desk quote more, more cheaply. Straight-through processing changes risk — it removes the re-keying errors and the reconciliation breaks. Digital onboarding changes both — it shortens the cash-to-first-trade cycle and improves the data quality that everything downstream depends on. A glossier client portal that does not change what the bank can price, how much it costs to serve a client, or which risks it carries is decoration, however impressive the demo. The direction of travel is real: standard flow becomes automated and self-service, and human sales moves up the value chain to ideas, complex business, and high-value relationships — the value migrating from execution to content, data, and technology. But the filter separates the trends that move the P&L from the ones that move only the screen.

The value migration, made concrete

Underneath the five trends is a single movement worth naming precisely: value is migrating from execution to content, data, and technology. In the old model, a bank was paid a spread or a commission largely for access — for being the pipe through which a client reached the market. Electronification is commoditising that pipe: when any client can reach any venue and put five banks in competition with one click, the pure execution of a standard trade is worth almost nothing. What is not commoditised is everything around the execution — the idea that made the trade worth doing, the data that told the bank which client wanted it, the liquidity the bank can provide in size when the market is thin, the analytics that prove best execution, and the technology that delivers all of it reliably. So the revenue does not disappear; it moves. The bank that will thrive is the one that runs the commoditised pipe superbly and cheaply while capturing the value that has migrated to the edges.

This reshapes the target operating model of a distribution business into a barbell, and the shape is worth drawing explicitly. At one end, a highly automated, low-cost, superbly engineered flow business handles the standard, liquid trades — priced by engine, executed by algorithm, settled by straight-through processing, and run at a cost-to-serve approaching zero. At the other end, a concentrated, high-touch, high-value business handles the complex, the illiquid, and the strategically important — priced by judgement, executed by a person, and justified by a relationship. The middle — the semi-standard flow once handled by a human at moderate cost — is where the pressure is fiercest, because automation is steadily eating into it from below. The strategic question every distribution head faces is how fast to push flow from the expensive middle to the cheap automated end without losing the relationships that the high-value end depends on. That single decision, repeated across every product and client, is what the transformation is actually about.

Generative AI in distribution, assessed honestly

Generative AI has real, emerging uses in distribution, and they cluster where language and data meet. LLMs can summarise research and long client chat histories so a salesperson walks into a call already briefed. They can draft client-facing content — market wraps, idea write-ups, meeting follow-ups — for a human to check and send. They sharpen communications surveillance, reading the huge volume of chat and voice for the patterns that matter and cutting the false positives that drown analysts. And they can narrate next-best-action, turning a model's suggestion into a readable rationale a salesperson will actually trust. Alongside these, the non-generative analytics that Part 2 introduced — axe-and-client matching, churn detection, win-rate pricing models — are the more mature half of the same story.

Use of GenAI The gain The governing risk
Summarise research and client chats Faster, better-briefed coverage Confidentiality; leakage across the wall
Draft client content Scale and speed in content production Hallucination; inaccurate market claims
Sharpen comms surveillance Fewer false positives, real risk found Explainability; model risk governance
Narrate next-best-action Suggestions salespeople trust and use Data privacy (GDPR); conflicts rules

The architecture of a governed GenAI assistant

To see why generative AI is both promising and hazardous in distribution, look at how a responsible assistant is actually built, because the architecture is the governance. A useful assistant does not simply send the bank's data to a model and relay the answer. It sits behind an entitlements gate that enforces exactly the information barrier of Part 3 — the model may retrieve only what the requesting user is entitled to see, so a public-side salesperson's assistant cannot reach private-side deal information. It uses retrieval-augmented generation, fetching the relevant, entitled documents and grounding the model's answer in them rather than trusting the model's memory, which is what keeps answers accurate and current. It passes the output through guardrails that check for leaked personal data, unsupported claims, and prohibited content. And it returns the result to a human who decides. Every one of those components is a control, which is why the architecture and the governance are the same thing.

Figure 19 A governed GenAI assistant: retrieval runs under an entitlements gate that enforces the information barrier, generation is grounded in retrieved documents and passed through guardrails, and a human makes the decision — with model risk governance wrapping the whole chain. Illustrative.

Model risk governance and the EU AI Act

Banks already know how to govern models, and generative AI is being pulled into that discipline rather than escaping it. Model risk governance is the established practice of validating, documenting, monitoring, and controlling the models a bank relies on — born in credit and market risk, now extended to the models that price, target, and surveil in distribution. Applied to generative AI, it asks the hard questions: what is this model for, what data trained it, how is it validated, how is its performance monitored, who is accountable when it errs, and can its decisions be explained? Explainability is the crux, because a surveillance or pricing model whose outputs cannot be explained fails the standard that supervisors — and increasingly the emerging EU AI Act, which classifies AI uses by risk and imposes obligations accordingly — demand. The near-term European reality is therefore adoption that is governed and explainable by design: efficiency and better targeting on top of clean data, with a human in the loop, rather than autonomous decision-making.

Delivering transformation: build, buy, or partner

The hardest question is not what to change but how to source it

Once a bank knows which transformation moves economics or risk, it faces a second decision that is just as consequential — whether to build it, buy it, or partner for it. The logic follows the buy-versus-build map from Part 4. The crown jewels — the pricing engine, the central risk book, the client-data platform, the decision logic that encodes the desk's edge — are built, because they are the competitive advantage and cannot be bought off a shelf without becoming identical to every competitor. The commodity utilities — market data, venue connectivity, the terminal — are bought, because building a worse copy of a standard wastes money. And the growing regulatory burden — surveillance, transaction reporting, TCA — increasingly leans toward specialist RegTech vendors and partnerships, because the rules change faster than an in-house team can keep pace and there is no competitive advantage in reporting the same trade the same way as everyone else.

The subtler point is that the sourcing decision is itself shaped by the architecture, which is why modernisation compounds. A bank with a clean, modular, API-connected estate can buy a best-in-class surveillance tool and plug it in, or partner with a fintech for a new analytics capability, in weeks — so it can assemble the best of build, buy, and partner. A bank whose systems are a tangle of point-to-point batch feeds must build everything itself or not at all, because nothing external will connect cleanly, and it falls further behind with each cycle. This is the deep reason the platform-modernisation programmes of Part 4 matter beyond their immediate payoff: they do not just improve today's systems, they buy the optionality to adopt tomorrow's innovations cheaply — whoever builds them. In a business being remade by technology, the ability to absorb the next technology quickly is itself the durable advantage.

The next frontier, assessed honestly: tokenisation and DLT

Beyond the five current trends sits an emerging one that deserves honest treatment rather than hype — the tokenisation of securities on distributed-ledger technology. The idea is that a bond or a fund unit could exist as a token on a shared ledger, settling near-instantly and atomically — delivery and payment in one indivisible step — rather than through the multi-day chain of depositories and custodians described in Part 2. If it worked at scale, it would compress the settlement cycle toward real time, reduce the reconciliation that the whole post-trade world exists to perform, and potentially let some instruments be distributed and traded on new rails entirely. Several central banks and exchanges are running live experiments, and the European regime has created a pilot framework to let regulated DLT market infrastructures operate under adapted rules.

The honest assessment applies the same filter as every other trend: does it change economics or risk, or just appearance? The genuine promise is real — atomic settlement removes settlement risk and reconciliation cost, which are exactly the kinds of change that move economics and risk rather than screens. But the obstacles are equally real: the existing infrastructure works, is deeply entrenched, and carries enormous value that will not migrate quickly; liquidity is hard to bootstrap on new rails; and the regulatory and legal questions about what a token legally is remain live. So the near-term reality is experimentation at the edges — specific instruments, specific corridors, controlled pilots — rather than a wholesale replatforming of distribution. For a technologist, tokenisation is worth watching closely and worth neither dismissing nor overselling: it is a credible long-run change to the settlement layer that will arrive, if it arrives, gradually and unevenly, one instrument and one corridor at a time.

  1. Name the six distribution risks and say which are least tractable to technology.

  2. Explain the win-rate fallacy.

  3. The three real cases each expose a different failure mode — name them and the mode.

  4. What single filter separates a real transformation trend from decoration?

  5. Under what conditions does generative AI earn its place in distribution, and why is the human in the loop a control rather than a limitation?

5

  1. What is the win-rate fallacy?

  2. Which distribution risks are least tractable to technology?

  3. Which case illustrates a market-liquidity collapse?

  4. Which trend changes risk rather than appearance?

The six ideas that matter most

Idea one: distribution is the bridge, and the channel follows the product.

It connects the bank's products and liquidity to the investors who buy them, and it assigns each product-and-client combination to the cheapest channel that still serves it well — voice for the bespoke and illiquid, API for the vanilla and liquid. The model is calibrated to the spectrum, not applied uniformly.

Idea two: data cascades, so the 360-degree view is the edge and its absence is the core failure.

A client interaction flows from CRM through pricing, execution, and booking to reporting and surveillance, and an error upstream reaches the regulator. The resolved client identity is the atom of the whole model, and fragmented client data — the reason there is no single view — is the pain point behind almost every other one.

Idea three: price discovery has moved to the platform and the engine.

For liquid flow the human has left the pricing loop; a pricing engine computes the mid, applies the spread, skews for the axe, checks the limits, and answers in milliseconds, while the protocol — RFQ, streaming, CLOB — decides where the dealer's edge lives. The price is made by the machine and the venue now, not the voice.

Idea four: compliance is core architecture, not an afterthought.

MiFID II, MAR, PRIIPs, GDPR, and the reporting standards are system specifications that reach into every block — classification in CRM, best-execution capture, communications surveillance, transaction reporting — and the FIX-to-ISO messaging seam and the resilience of the critical moment are where those specifications are met or missed.

Idea five: wallet share and cost-to-serve quietly distort behaviour.

The competition is for a standing slice of each client's spend, not for any single trade, and the economics — revenue from wallet share against the cost of covering a client — decide where human coverage and technology go. It is the league table of distribution, and like every scoreboard it shapes what people do.

Idea six: real transformation moves cost or risk, including generative AI under governance.

Auto-pricing, electronic protocols, single-dealer platforms, digital onboarding, and straight-through processing rebuild the model, and each is worth judging by whether it changes economics or risk. Generative AI belongs on that list only where it does the same, on clean data and under tight control, with the human in the loop as the control itself.

Process Cheat Sheet — The Whole Model on One Page

A single reference for the entire distribution model — the workflow, the channels, and the numbers that frame it. Every figure is a rule of thumb or illustrative, not a quoted fact.

The workflow at a glance

Step Timing Core activity Key output Lead and systems
1. Content & idea Continuous Absorb research, colour, axes; form a view A relevant, sized idea Sales, research; CRM, next-best-action
2. Reach out / pitch Continuous Contact the right clients on the right channel An engaged client Sales, e-commerce; chat, SDP, axe broadcast
3. Enquiry / RFQ Seconds–minutes Client asks for a price or shows interest A live request Client; MDP, SDP, RFQ
4. Price & negotiate Milliseconds–minutes Compute mid, spread, skew; quote A firm quote Desk, pricing engine; e-pricing
5. Execute & confirm Instant Deal done, captured, confirmed An executed trade Sales-trading; OMS/EMS, FIX
6. Allocate & settle T+1/T+2 Allocate blocks/new issues; clear and settle A settled position Syndicate, ops; bookbuilding tools, STP
7. Service & follow-up Continuous Reporting, best-ex evidence, the next idea A renewed relationship Sales, controls; TCA, reporting, analytics

The channel spectrum

Channel Touch Best for Margin
Voice Highest Complex, large, illiquid trades and ideas Highest
Chat / IM Everyday Fast, logged, semi-structured relationship flow High
Single-dealer platform Owned High-value clients, differentiated content Higher
Multi-dealer platform Competitive Standard flow where clients compare banks Compressed
Direct API / algo Lowest Vanilla, liquid, systematic flow Thinnest

Key numbers — rules of thumb (illustrative)

Metric Rule of thumb
FX electronic share ~80%+ — the most automated market
Cash equities electronic share ~90%+ — order-book and algo dominated
Rates electronic share ~60–70% and rising
Credit electronic share ~40–50% and rising fastest
Structured products electronic share Low — largely human and advisory
New-issue oversubscription (healthy) ~2–3x covered before tightening
Auto-quote response time Milliseconds — a slow price loses the trade
Settlement cycle T+1 to T+2 depending on market
Wallet-share goal per key client Top-three in the client's wallet
Onboarding time (pre-digitisation) Often weeks — the friction to attack

Key systems and vendors at a glance

Layer What it does Typical names Build or buy
Terminal / data hub Market data, chat, RFQ flow Bloomberg Buy
Multi-dealer platforms Competitive RFQ, portfolio trading Tradeweb, MarketAxess, 360T Connect
Pricing engine Auto-quote, skew, risk check Proprietary Build
Central risk book Net and internalise flow, hedge residual Proprietary Build
CRM / client-data platform 360 view, entity resolution Vendor + proprietary Mix
OMS / EMS / SOR Order routing, execution, best-ex Vendor or in-house Mix
Surveillance & reporting Comms/trade surveillance, TCA, txn reporting Specialist RegTech Buy

Regulation quick-reference

Regime What it governs Where it bites
MiFID II / MiFIR Conduct, transparency, best-ex, reporting Every block; the master spec
MAR Market abuse Surveillance, the information barrier
PRIIPs Packaged retail products KID for structured products
GDPR Personal data CRM and analytics
EMIR / SFTR Derivatives / financing reporting Trade-repository reporting
CSDR Settlement discipline Penalties for failed settlement
DORA Operational resilience Resilience testing, third-party oversight

The whole model in one sentence

Distribution is the machine that connects a bank's products and liquidity to the investors who buy them — matching the right idea and price to the right client, through the cheapest channel that still serves them well, and proving to the regulator at every step that the client was treated fairly.