Two threads run through this module
The business thread follows the view from idea to action. An analyst spots a mispricing, builds the case, sets a rating and a price target, passes a hard compliance gate, and publishes to every entitled client at the same instant. Sales and advisers then carry that view to two very different audiences: institutions who decide for themselves, and wealth clients who must be given advice that is demonstrably suitable for them. The through-line is that research only has value if it is independent, differentiated, and provably consumed.
The technology thread follows the same view through the systems that carry it. A research note is authored in one platform, reviewed in a workflow, published once under a web of entitlements, distributed simultaneously across terminals and portals, and then tracked so the bank knows who read it and can charge for it. Since research was forced to be paid for separately, the system that measures consumption became as important as the system that produces the note. Both threads meet at the same place: independence and fair distribution are not policy overlays but the architecture of the whole business.
How to read this module
Read it front to back the first time. Each part builds on the last: the product before the process, the process before the people, the people before the platform, the platform before the risks. After that, treat it as a reference — the cheat sheet and glossary at the end are built for lookup.
Throughout, coloured boxes carry specific kinds of information. A blue Key insight box crystallises an idea worth remembering. A navy Going deeper box adds detail for readers who want the mechanics. A teal IT & data lens box shows how a concept lands in systems and data. An orange Watch out box flags a trap or a live risk. A green Worked example box follows our running fictional case. A light-blue In a nutshell box recaps each part, and a purple Check your understanding box closes it with questions and answers. One fictional case — a European mid-cap equity initiation — runs through the whole module so that every abstract mechanic has something concrete to hold onto.
What you will master
You will understand what research is and why its independence is the entire product
You will be able to explain, from first principles, what a research note contains, how a rating and price target are built, and why an analyst who can be leaned on is worth nothing to a client and dangerous to the bank.
You will be able to trace a view from idea to a client acting on it
You will follow the seven-step production pipeline end to end, know where the hard compliance gate sits, and understand why a note cannot reach one client a second before another.
You will be able to map every actor and where the walls run between them
You will know who produces research, who distributes it, who consumes it, and why an information barrier has to cut through the middle of the same institution.
You will understand the systems that author, publish, entitle and measure research
You will be able to describe the research platform end to end — authoring, review, entitlements, distribution and consumption analytics — and explain why measuring consumption became a business-critical capability overnight.
You will recognise the shock that reshaped the business and the transformation now under way
You will understand the MiFID II unbundling revolution and its partial reversal, the conduct rules that govern every note, the suitability duty that attaches to personal advice, and how data and generative AI are changing what an analyst does.
1Context and the Big Picture
Before the process, the people, and the platform, one question has to be settled: what is research actually for, and why does a bank spend a fortune producing it? This part builds the answer from zero — what a view is, why it has value, and how one regulation turned a hidden cost into a business that must justify every euro.
A research view is a defensible answer to the three questions every investor asks
Every investor, professional or not, is really asking three questions, and research exists to answer them. What is this security actually worth, and why? What is happening in this market or economy? And, given all of that, what should I do with my money? Answer the first well and you have equity or credit research. Answer the second and you have macro and strategy. Answer the third for a specific person and you have investment advice. The whole discipline is the manufacture of informed, defensible answers to those three questions, produced at industrial scale and sold to people who could not easily produce them themselves.
A "view" sounds soft, but it is a precise object. It is a claim about the future — this stock will outperform its sector over twelve months, this currency will weaken against the dollar, this issuer's spread will tighten — attached to an argument, a number, and a named person who will be held accountable if it is wrong. The claim without the argument is a guess. The argument without the number is a lecture. Research is all three at once, which is why it is expensive: it takes a trained analyst weeks to build a view that survives contact with a sceptical client.
The reason a bank produces this at all is that views drive flow. A client who trusts a bank's research picks up the phone to that bank, trades through its desks, attends its conferences, and takes its calls. Research is the content that makes a bank worth engaging with — the visible expression of the franchise. It rarely earns its keep as a standalone product; it earns it by pulling everything else along behind it. Hold that idea, because in 2018 a single regulation forced research to earn its keep directly, and the business has never been the same since.
Research is general; advice is personal, and the line between them is legal
The single most important distinction in this module is the one between research and advice. Research is general: a view published to many clients at once — "this stock is a Buy, target 120." Investment advice is personal: a recommendation tailored to one client's specific situation — "given your portfolio and your objectives, you should reduce this position." The two look similar and are often produced by the same institution from the same underlying analysis, but the law treats them completely differently.
The moment a recommendation becomes personal, a duty of care attaches that general research never triggers. Personal advice carries a suitability obligation: the firm must ensure the recommendation genuinely fits the client's knowledge, financial situation and objectives, and must be able to prove it did. General research carries no such duty — it is addressed to the market, not to you. This is not a semantic nicety. It is the difference between a business that publishes and a business that advises, and it shapes the systems, the controls, and the headcount of each. We return to suitability in depth in Part 2 and Part 4; for now, hold the line firmly: general versus personal is a legal boundary, not a matter of tone.
Research sits upstream of every trade, at arm's length from the rest of the bank
Picture the markets business as a chain, and research sits at the head of it. Raw material flows in at one end — company filings, economic data, market prices, and increasingly alternative data such as card spending and satellite imagery. Research turns that raw material into views, ratings and forecasts. Sales and advisory distribute those views and advise clients on what to do. Buy-side clients consume the views, form decisions, and trade. And the resulting orders execute through the desks. Research is at the top of the waterfall: it shapes what clients think, and therefore what they do, and therefore where the flow goes.
Because research sits upstream, its independence is not a nicety — it is the whole value. If an analyst's view can be bent by the bank's other interests, the view is worthless to the client and dangerous to the bank. So research sits behind an information barrier: walled off from investment banking, and partly separated from the sales-and-trading desks whose positions its ratings could move. That arm's length is what a client is really paying for. A compromised analyst produces content that looks like research but functions as marketing, and the market eventually notices. Part 2 shows how the wall is built and enforced; the point to fix now is that independence is the product, not a constraint on it.
Three kinds of institution produce research, and this module is about one of them
Not all research comes from banks, and the differences matter commercially. Three distinct types of institution produce it, each with a different conflict profile and a different reason to exist.
| Producer | Who they are | Why they produce it | Conflict profile |
|---|---|---|---|
| Sell-side research | Banks and brokers | To win and keep clients' trading business — the focus of this module | Highest: sits inside an institution that also does banking and trading |
| Buy-side research | Asset managers, hedge funds | For their own investment decisions — proprietary, not distributed | Internal only: never sold, so no distribution conflict |
| Independent research providers (IRPs) | Boutiques and specialists | Sold directly to investors for a fee | Lowest: no banking or trading business to corrupt the view |
Sell-side research is the subject of this module, and it carries the heaviest conflict burden precisely because it lives inside a full-service bank. The same institution that publishes a Buy rating on a company may also want that company's next bond mandate, and may hold a position in the stock on its trading book. Managing those conflicts is the central design problem of the whole business, and it is why sell-side research is the most heavily regulated of the three. Buy-side research never leaves the building, so it has no distribution or conflict issue. IRPs sell a clean conscience — no banking, no trading, no axe to grind — which, as we will see, became a competitive advantage the moment clients had to pay for research explicitly.
Advice comes on a spectrum, from no advice at all to full discretion
Between "here is a view, do what you like" and "we will manage your money for you" lies a spectrum of responsibility. Understanding where each service sits on that spectrum tells you how heavily it is regulated, because responsibility to the client rises as you move along it.
| Service | What it means | Advice given? | Regulatory weight |
|---|---|---|---|
| Execution-only | The client decides; the bank just executes | None | Lightest |
| General research | Views published to many; not personal advice | No personal advice | Conflicts and fair-presentation rules |
| Investment advice | A personal recommendation to a specific client | Yes, personal | Suitability applies |
| Discretionary management | The bank decides and trades for the client | Yes, and acts on it | Heaviest |
The sell-side markets business lives mostly at the left of this spectrum: general research plus execution. It publishes views and executes trades, but rarely gives personal advice to institutional clients, who are presumed sophisticated enough to decide for themselves. Personal advice concentrates at the right of the spectrum, in wealth management and private banking, where clients want recommendations tailored to their circumstances and the firm takes on a suitability duty for each one. The same house research can feed both journeys — an institution acts on it directly, a wealth adviser translates it into a suitable personal recommendation — but the regulatory obligations differ completely between the two.
Europe pushed research into a corner MiFID II created, and France sits at its centre
In Europe, two of the four services above are formally regulated investment services, and that shapes everything. Under MiFID II — the EU framework governing markets conduct — investment advice and portfolio management are regulated services with strict suitability rules, while research is separately regulated for independence and conflicts. This is stricter and more explicit than the equivalent US regime, and it means European research businesses carry a heavier compliance and systems burden than their American counterparts. National regulators enforce it: in France, the Autorité des marchés financiers (AMF) supervises research independence and advisory conduct directly.
Europe also did something the United States never did: it unbundled research from trading, then partly reversed course. That single reform, which we unpack at the end of this part and again in Part 5, is the defining event of the modern research business, and it is a European story with no direct US equivalent. The EU has since acted to revive sell-side coverage of smaller companies through the Listing Act, precisely because unbundling had hollowed it out. If there is one reason this module carries a European and French lens throughout, it is that the rules that reshaped research were written in Brussels, refined in Paris and London, and are still being rewritten today.
Five research desks cover every market, but they all do the same job
Research is not one activity but a family of them, split by the market each desk covers. The craft is identical across all of them — gather data, build a rigorous analytical view, and communicate it so a client can act — but the raw material and the output differ. Five desks cover essentially every tradable market.
| Desk | What it covers | Core output | Who consumes it |
|---|---|---|---|
| Equity research | Listed companies, by sector | Earnings models, valuations, ratings, price targets | Equity investors, the classic franchise |
| Credit research | Bond issuers' creditworthiness | Default risk, spread views, recovery analysis | Credit desks, fixed-income investors |
| Macro & economics | Growth, inflation, central-bank policy | Forecasts and the big-picture context | Everyone; it moves all markets |
| Rates & FX strategy | Bond yields, curves, currencies | Actionable market trade calls | Rates and FX desks and investors |
| Commodities & other | Oil, gas, metals, agriculture | Supply/demand analysis | Producers, consumers, investors |
Equity research is the biggest and most visible desk, which is why it anchors this module. It is headcount-heavy, sector-organised, and the most familiar face of the business — the Buy and Sell ratings that reach the financial press come from here. But visibility is not the same as impact. A single macro call on the direction of rates can move more money than a hundred equity ratings, because it reprices entire asset classes at once. Alongside the five core desks sit fast-growing, data-driven areas — quantitative and factor research, cross-asset thematics, and sustainability (ESG) analysis — which increasingly cut across the traditional desks rather than sitting beside them.
Why we use equity as the reference case
Equity research is to this module what the IPO is to equity capital markets: the most complete, most visible, most headcount-intensive version of the craft, and therefore the best teaching case. It has the fullest anatomy — models, valuations, ratings, targets, corporate access — and the sharpest conflict profile, because equity analysts sit closest to the investment-banking deals whose companies they cover. Master equity research and the credit, macro and strategy variants become straightforward adjustments to the same template.
A research note is a differentiated argument wrapped in mandatory disclosure
The core product of the business is the note, and every note has the same anatomy for a reason. A busy client reads it in a particular order, and each part exists to serve a different need — the client's need to act, the analyst's need to persuade, and the bank's need to stay on the right side of the regulator. Understand the six parts and you understand what research actually sells.
| Part of the note | What it contains | Why it is there |
|---|---|---|
| Headline & rating | The call in one line: "Buy, target 120," and the change | What a busy client reads first |
| Investment thesis | The core argument for why the security will move | The differentiated view that adds the value |
| Financial model | Earnings forecasts, valuation, the numbers behind the target | The analytical backbone |
| Catalysts & timeline | What events will move the price, and when | So the client knows what to watch and when to act |
| Risks | What could make the thesis wrong | Honest analysis and a compliance requirement at once |
| Disclosures | Conflicts, relationships, rating history, methodology | Mandatory and heavily regulated |
The thesis is where the value lives, and it is the hardest part to produce. Anyone can pull a consensus estimate off a terminal; the reason a client pays is the differentiated argument — the reason this analyst thinks the market is wrong, laid out well enough to act on. The model is the backbone that makes the thesis credible, the catalysts tell the client when to act, and the risks section is both good analysis and a legal requirement. The disclosures are the part nobody reads and everybody needs: they reveal the bank's relationships with the company, the analyst's own holdings, and the distribution of the bank's ratings, so the reader can judge how much to trust the call.
A rating turns a view into a standardised, accountable signal
A view expressed in prose cannot be compared, tracked, or fed to a machine; a rating can. So every bank compresses the analyst's argument into a standardised signal. Most use a three-tier stock rating, though the labels vary: Buy, Outperform or Overweight for a stock expected to beat the market; Hold, Neutral or Market-weight for one expected to perform roughly in line; and Sell, Underperform or Underweight for one expected to lag. Alongside the rating sits a price target — a specific level on a stated horizon, usually twelve months — and often a note on whether the rating is absolute or relative to the sector or index. Banks also maintain conviction lists: the handful of ideas the analyst or the firm backs most strongly.
Formalising the call into a rating serves four distinct purposes, and only one of them is about the client. A standard scale lets clients compare ideas across analysts and banks quickly. A rating and a target create a track record, so analysts can be measured on whether their calls were right — reputation is an analyst's currency. Structured ratings and targets feed platforms and alerts, so clients see changes the instant they publish. And regulation requires the definitions, the distribution of ratings, and their histories to be disclosed, so the signal cannot quietly mislead.
The Buy-rating bias, and the rule written to fight it
Historically, banks published far more Buy ratings than Sell ratings, and the reason was a conflict: a Sell rating angers a company the bank might want as an investment-banking client, and cuts off the analyst's access to management. After the analyst scandals of the early 2000s — which we examine as a case study in Part 5 — regulators forced banks to disclose the distribution of their ratings, so a client can see what percentage are Buy versus Sell, and any banking relationship with a rated company. In Europe this sits inside MiFID II's conflicts rules and MAR's requirement that investment recommendations be fairly presented. The rule does not ban optimism; it makes systematic optimism visible.
The business is large, concentrated, and now measured to the euro
Research has always been a big number hidden inside a bigger one, and unbundling dragged it into the light. Global spending on sell-side research runs into the billions of dollars a year, but for decades no one could state it precisely, because it lived inside trading commissions. Since 2018 the figure has been explicit, contested, and falling. The shape of the business matters as much as its size: spending concentrates on a handful of top providers, a small number of star analysts command outsized client attention, and the highest-value services — analyst and corporate access — are worth more to many clients than the written note. The table below frames the business with illustrative orders of magnitude; the exact figures move each year, but the proportions are the point.
| Dimension | Illustrative shape of the business |
|---|---|
| Spend concentration | The buy-side pays a small number of providers most of its research budget |
| Coverage universe | A lead analyst covers ~15-25 names; a bank covers hundreds across desks |
| Highest-value service | Analyst and corporate access, often rated above the written note |
| Budget trend since 2018 | Sharply down, then stabilising, with a partial reversal for SMEs |
| Ranking currency | Buy-side surveys (e.g. Institutional Investor) drive an analyst's reputation |
The reputational economy runs on client votes, and that is a genuinely unusual market structure. Analysts are ranked by the clients they serve, through surveys in which the buy-side votes on who adds the most value. A top-ranked analyst pulls client attention, trading flow and, ultimately, budget toward the bank; a slip in the rankings does the opposite. This means an analyst's individual reputation is a bankable asset that can move between employers, which is why star analysts are fought over and why a research franchise is more fragile than an org chart suggests. The value walks out of the building every evening.
Quant, thematic and ESG research are the fastest-growing corner of the business
The traditional five desks are being cut across by three data-driven styles that do not respect sector boundaries. Quantitative and systematic research uses large datasets and statistical models to find signals across whole markets rather than forming a view on one company; its output is often a factor or a screen rather than a note. Thematic research follows an idea — energy transition, ageing demographics, artificial intelligence — across every asset class it touches, cutting horizontally through the vertical sector desks. And sustainability research analyses emissions, taxonomy alignment and governance, feeding a fast-growing client demand that European regulation has made close to mandatory.
These styles grew because the buy-side changed what it wants, and Europe's rules accelerated the sustainability strand in particular. Quant clients want structured data, not prose, which rewards banks that can deliver machine-readable output. Thematic research sells because a single big idea gives many clients a reason to act at once and showcases the whole franchise. And ESG research is propelled by the EU's sustainability-disclosure regime, which turns emissions and taxonomy data into required inputs rather than optional colour — making sustainability analysis both a research growth area and a data-engineering challenge in its own right.
How research is priced now, having once been free
The deepest business change unbundling forced was not smaller budgets but the invention of a price for something that never had one. Before 2018 there was no market price for a research note, an analyst call, or a management meeting — they were all bundled into commission. Overnight, the whole industry had to invent pricing, and it settled on a small set of models that now coexist. A bank funds its own research through its profit and loss or through a client-funded Research Payment Account; a client, in turn, is charged through a mix of models depending on what it consumes.
| Pricing model | What the client pays for | Best suited to |
|---|---|---|
| Tiered subscription | Flat-fee access at a service level (read-only to full interaction) | Broad, predictable consumption |
| À la carte | Analyst time, bespoke projects, corporate access, priced individually | High-value, occasional services |
| Consumption-based | A fee negotiated from measured readership and interaction | Data-rich relationships |
| P&L-funded (bank pays) | Nothing directly; the bank absorbs the cost | Franchise support, flow attribution |
Pricing research is genuinely hard because value and volume are unrelated, and the market solved it with data plus dialogue. One brilliant call can be worth more than a thousand routine notes, and the same note is priceless to one fund and useless to another — so a price cannot be read off a page count. What the market converged on is a two-part answer: measure consumption objectively, then negotiate an annual value in a dialogue that is part data and part relationship. Tiered subscriptions simplify the base; à la carte pricing captures the premium services — analyst access, bespoke work, corporate access — that clients value most; and the consumption evidence anchors the negotiation so neither side is guessing. The uncomfortable truth underneath is that pricing research is as much a commercial-relationship skill as an analytical one.
The advisory business earns its keep in a completely different way
Research sells views; advisory sells outcomes, and the two are paid for through entirely different mechanisms. Sell-side research is monetised indirectly — through the trading flow it drives and, since unbundling, through explicit research fees. Advisory, which lives mostly in wealth management and private banking, is monetised directly, through fees charged to the client for the advice itself. Understanding these revenue models matters, because they explain why the advisory business is structured, staffed and regulated so differently from research.
| Advisory revenue model | How it is charged | What it aligns |
|---|---|---|
| Advisory fee | A charge for personal recommendations | Advice quality with client relationship |
| Discretionary management fee | A percentage of assets managed | The bank's revenue with portfolio size |
| Product & transaction charges | Fees on products bought or trades executed | Historically prone to conflict |
| Retrocessions (inducements) | Payments from product providers to the adviser | Restricted or banned under MiFID II |
The most consequential regulatory change in advisory economics is the assault on retrocessions, and it reshaped the European wealth business. A retrocession is a payment a product provider makes to the adviser who sells its product — a structure that quietly incentivises the adviser to recommend the products that pay best rather than those that suit the client best. MiFID II restricts these inducements sharply, banning them outright for independent advice and discretionary management, which pushed European wealth managers toward transparent, fee-based models. This is why advisory has moved from a hidden-commission business toward an explicit advice-fee business — the same journey research made with unbundling, running in parallel.
Advisory is a large and distinctly European business, and France sits close to its centre. The private-banking pools of France, Switzerland and Luxembourg make wealth advisory a substantial revenue line, not an afterthought to the markets business. It is also the most heavily protected corner of the client world: retail and wealth clients receive suitability, appropriateness and product-governance protections that institutional clients never do, supervised in France by the AMF. The result is a business that trades scale for margin — fewer clients than research reaches, but each one a documented, fee-paying, long-term relationship rather than a recipient of a broadcast note.
What is the legal difference between general research and personal investment advice?
General research is published to many clients and triggers no personal-advice duty; personal advice is tailored to a specific client and triggers a suitability obligation to ensure it fits that client's situation.
Why is a research analyst's independence described as "the whole product"?
Because a view that can be bent by the bank's banking or trading interests is worthless to a client and dangerous to the bank; clients pay precisely for a view they can trust and cannot easily produce themselves.
Name the three kinds of institution that produce research and rank them by conflict burden.
Sell-side (banks/brokers, highest conflict), buy-side (asset managers, internal only), and independent research providers (lowest, no banking or trading business).
What are the six parts of a research note, and which one carries the value?
Headline & rating, investment thesis, financial model, catalysts & timeline, risks, and disclosures. The differentiated investment thesis is where the value lives.
Why do banks compress a view into a standardised rating and target?
For comparability across analysts, accountability via a track record, machine distribution and alerts, and regulatory disclosure of rating definitions and distributions.
What did MiFID II unbundling change in 2018, and why?
It forced research to be paid for separately from trading rather than bundled into commissions, to remove the conflict where "free" research was an inducement to over-trade and to make research's cost transparent.
How have Europe and the UK partly reversed unbundling, and why?
The EU Listing Act re-permits bundled payment and removes the market-cap threshold from 2026, and the UK FCA introduced a joint-payment option from 2024, both to revive research coverage of smaller companies that unbundling had hollowed out.
What pricing models replaced free research, and what ties them together?
Tiered subscriptions, à la carte pricing of premium services (analyst time, bespoke work, corporate access), consumption-based fees, and P&L funding by the bank — tied together by an annual value dialogue that sets each client's fee from measured consumption.
How does the advisory business earn revenue differently from research, and what changed it most?
Advisory charges the client directly through advice and discretionary-management fees rather than being monetised through flow; MiFID II's restriction of retrocessions (inducements) pushed European wealth management toward transparent, fee-based models.
1
What separates general research from personal investment advice?
Format is not the test.
Price is not the legal distinction.
The suitability obligation is the difference.
Right. General research goes to many clients and does not.
Why is an analyst's independence called “the whole product”?
Right. Clients pay for a view they can trust.
Not the reason.
No.
They work in teams.
Which research producer carries the heaviest conflict burden?
Its research is internal.
Lowest conflict: no banking or trading.
Right. They also bank and trade with the companies covered.
Not part of the taxonomy.
Which part of a research note carries the value?
Mandatory, but not where the value is.
No.
The rating summarises the thesis.
Right.
2The End-to-End Production Flow
A capital markets deal has a linear life: mandate, preparation, marketing, pricing, settlement. Research has one too, and it is just as disciplined, though what it produces is an argument rather than a security. This part follows a single view — our fictional Meridian initiation — from the spark of an idea to the moment a client acts on it, stopping at every mechanic that matters: the valuation maths, the hard compliance gate, the fair-distribution rule, and the two very different journeys the same view takes to institutional and wealth clients.
Hold the whole pipeline in your head before the detail
Every research note, from a two-line flash to a fifty-page initiation, passes through the same seven stages. The value is created in the first three, the risk is controlled in the fourth and fifth, and the money is captured in the last two. Seeing the whole arc first makes each stage easier to place.
| Stage | What happens | Typical duration | Who leads |
|---|---|---|---|
| 1. Idea & coverage | The analyst spots something worth saying | Ongoing | Lead analyst |
| 2. Analysis & modelling | Gather data, build the model, set rating and target | Days to weeks | Analyst + associate |
| 3. Draft the note | Write the argument: thesis, model, catalysts, risks | Days | Analyst + associate |
| 4. Supervisory & compliance review | Check accuracy, balance, disclosures, conflicts | Hours to days | Supervisory analyst + compliance |
| 5. Publish | Release simultaneously to all entitled clients | Minutes | Publishing / entitlements system |
| 6. Distribute & market | Push to platforms; analyst and sales market the call | Days to weeks | Research sales + analyst |
| 7. Maintain & follow | Update on events; track the call's performance | Ongoing | Lead analyst |
Effort and risk are not spread evenly across the seven stages — they concentrate at two points. The analytical effort peaks in stages two and three, when the view is built and written. The compliance risk peaks sharply at stages four and five, the review-and-publish moment, because that is where a conflicted, unbalanced, or selectively-distributed note does its damage. Everything before stage four is reversible; a draft can be rewritten. Once a note publishes to the market in stage five, it cannot be unpublished, and any breach is now in the wild. That asymmetry is why the technology and the controls cluster around the publish moment.
Step 1 — Coverage is a franchise decision before it is an analytical one
Before an analyst can have a view on Meridian, the bank must decide Meridian is worth covering at all, and after 2018 that is a hard commercial choice. A lead analyst "covers" a defined universe — typically fifteen to twenty-five stocks in a sector, or a market. Adding a name to that universe commits scarce, expensive analyst time to it indefinitely. In the pre-unbundling world, coverage was cheap to add and rarely dropped; research was free, so breadth was a virtue. After unbundling, every name in the universe has to earn its place, because the bank is now paying visibly for the analyst's time. Covering an under-researched mid-cap like Meridian is a bet that a differentiated view will attract enough client interest and trading to justify the cost.
The idea itself is the analyst's raw material, and good ideas share a shape. They identify something the market is mispricing — an event, a structural change, a theme — that the analyst can argue better than anyone else. A "me-too" view that agrees with consensus is worthless, because the client can get it free from a dozen other banks. The whole game is the variant view: a defensible reason to believe the market is wrong. Meridian qualifies precisely because it is under-covered and the margin inflection is not yet in consensus.
Independence is established before the first number is written
The moment an analyst begins forming a view, the machinery protecting its independence is already running. This is the research equivalent of a bank committing its balance sheet: the point at which the institution's other interests could contaminate the product, and therefore the point at which the walls must already be up. Four conflicts have to be held off from the start. Investment banking may want a favourable view on a company it does deals for, so research is walled off from the bankers. The trading desk may hold a position a rating would help or hurt, so research must not be steered by the book. Analysts must never publish or trade on material non-public information. And an analyst's own holdings are restricted, pre-cleared and disclosed. Part 3 details how the wall is built; the point here is that it is load-bearing from step one, not bolted on at the end.
Independence is enforced in systems long before it appears in a note. Before our analyst can even open a model on Meridian, the compliance platform checks whether the stock is on a restricted list — meaning the bank is involved in a transaction that bars research from commenting. It logs any interaction between research and the private side of the bank as a controlled wall-crossing. And it monitors the analyst's personal-account dealing to ensure she is not trading ahead of her own view. The wall is a policy, but it lives as access controls, restricted-list feeds, and surveillance in the research and compliance platforms. A wall that exists only on paper is not a wall.
Step 2 — The view is built in a model, and the target is triangulated, not guessed
The analytical heart of the process is the financial model, and its output is a defensible number. With the associate's support, the analyst gathers the data — filings, transcripts, channel checks, market prices — and builds an earnings model that forecasts the company's revenue, margins and profits over several years. The model is not the view; it is the machine that turns the view into numbers a client can interrogate. If the analyst believes Meridian's margins will inflect, the model has to show exactly how much, when, and what that does to earnings and to the share price. A thesis a client cannot pressure-test in the numbers is not research; it is an opinion.
A price target is built from at least two independent valuation methods, and the honest answer is a range, not a point. No single method is trusted on its own, so the analyst triangulates. The two workhorses are a discounted cash flow, which values the company on the cash it will generate, and a multiples comparison, which values it against what similar listed companies trade for. Where the two methods converge, the analyst has a defensible range; where they diverge, the analyst has to explain why. The target price is a judgement drawn from that range, not a mechanical output.
A worked valuation for Meridian
Take the illustrative numbers. Meridian trades at 20 per share today, with 100 million shares, so its equity is worth 2.0 billion. The analyst's discounted cash flow, built on the margin-recovery forecast, values the equity at about 2.6 billion, or 26 per share. A multiples comparison — applying the sector's average enterprise-value-to-EBITDA multiple to Meridian's forecast EBITDA — gives a range of 2.4 to 2.8 billion, or roughly 24 to 28 per share. A third check, recent precedent transactions in the sector, points to 25 to 30 per share but is given less weight because takeovers include a control premium. The methods converge around 26, so the analyst sets a twelve-month target of 26 — thirty percent above the current 20 — and rates the stock a Buy. The convergence is what makes the target defensible; a single method could be dismissed as cherry-picked.
Why insist on triangulation when a discounted cash flow alone produces a precise number? Because precision is not accuracy. A discounted cash flow is exquisitely sensitive to its assumptions — a small change in the discount rate or the terminal growth rate can move the value by a third. Multiples anchor that abstraction to what the market is actually paying for comparable businesses right now. Precedent transactions add a third reference point but must be discounted for the control premium a buyer pays to own the whole company. The convergence of independent methods is what lets an analyst defend a target to a sceptical client — and to a compliance officer asking whether the number is supportable. A target that rests on one method is a target that rests on one set of assumptions.
The Meridian discounted cash flow, in illustrative outline. The analyst forecasts free cash flow rising as margins recover — say from around 180m this year toward 260m in five years as the restructuring matures. She discounts those flows at a weighted average cost of capital of roughly 8.5%, and adds a terminal value capturing the cash beyond the forecast, grown at a long-run rate of about 2%. Summed and discounted to today, the enterprise value comes to roughly 2.8bn; subtract net debt of around 0.2bn and the equity is worth about 2.6bn, or 26 per share on 100m shares. The sensitivity is the lesson: nudge the discount rate to 9.5% and the value falls toward 23; nudge the terminal growth to 3% and it rises past 28. That swing — from a single assumption — is exactly why the analyst refuses to trust the DCF alone and triangulates it against multiples. All figures illustrative.
The rating is calibrated against consensus, because the value is in the difference
A rating only means something relative to what the market already believes, so the analyst's real product is the gap. Consensus — the aggregated forecasts of all analysts covering a stock — is the reference point. If the analyst's earnings forecast for Meridian sits comfortably above consensus, she has a "variant view," and the Buy rating expresses it. If her numbers matched consensus, the Buy would add nothing, because the market already agrees. This is the research equivalent of price discovery: the analyst is not setting a price in a vacuum but locating her view relative to the collective view, and selling the distance between them. The bigger and more defensible the gap, the more valuable the call — and the more it will move the stock if the market comes around.
Step 3 — The note turns the model into a persuasive, defensible argument
With the numbers settled, the analyst writes the note, and writing is where analysis becomes communication. The associate drafts sections and runs the numbers; the analyst owns the thesis and the call. The note follows the anatomy from Part 1: the headline and rating first, then the investment thesis that carries the argument, the model that backs it, the catalysts that tell the client when to act, the risks that could break the thesis, and the disclosures. For Meridian, the thesis is the margin inflection the market is missing; the catalysts are the next two earnings releases where the inflection should become visible; the risks are that the restructuring stalls or that end-markets weaken. A good note makes the reader feel the conviction and see the caveats at once.
The hardest discipline in drafting is honesty about what could go wrong. The risks section is not a compliance formality bolted to the end; it is where a good analyst demonstrates she has stress-tested her own view. Naming the ways the thesis could fail makes the note more persuasive, not less, because it signals the analyst has already considered the objections a sceptical client will raise. It is also a hard regulatory requirement: a recommendation that hides its risks is not fairly presented.
Step 4 — The compliance gate is where the note becomes irreversible
Before a note reaches a single client, it passes a review that is the research equivalent of a bank's commitment committee. Two checks run in parallel. A supervisory analyst — a specialist reviewer, a required control role — checks the note for analytical accuracy, internal consistency, and balance: do the numbers add up, does the rating match the target, is the risk disclosure adequate. Compliance checks conflicts and disclosures: is the bank in a transaction with Meridian that bars the note, are all relationships disclosed, is the analyst's personal position clean, is the stock off the restricted list. Only when both sign off can the note publish. This is the hard gate, and it is deliberately a bottleneck.
The gate exists because after it, nothing can be taken back. A draft is infinitely revisable; a published note is a fact in the market. If a conflicted or unbalanced note gets through, the damage is done the instant it reaches clients, and no correction fully undoes it. So the review is where much of the research business's control effort — and much of its technology — concentrates: authoring tools that assemble disclosures automatically, workflow systems that enforce sign-off before release, and audit trails that prove who approved what and when. The gate is slow on purpose, because the cost of a bad note escaping is far higher than the cost of an hour's delay.
The internal gauntlet: restricted lists, quiet periods, and conflict clearance
Three specific controls sit inside the gate, and each closes a specific loophole. The restricted list bars research from commenting on any company where the bank has material non-public information or an active transaction — if the bankers are working on a Meridian deal, the analyst cannot publish, full stop. Quiet periods restrict research around deals the bank is involved in, such as an IPO it is underwriting, to prevent research being used as promotional cover. And conflict clearance requires every note's disclosures to reflect the bank's actual relationships with the company, checked against internal records rather than the analyst's memory. Together they form an internal gauntlet the note must clear before it is allowed out — the equivalent of the internal committees a capital markets deal must pass before the bank commits.
Step 5 — Fair, simultaneous distribution is a hard line, not a courtesy
The single most important rule in distribution is that every entitled client receives the note at the same instant. A note cannot go to a favoured client a minute before others. The moment research reaches one client ahead of the rest, that client has an informational edge the others were denied, and the bank has handed out what amounts to a selective tip. So publication is a controlled, simultaneous release: the note goes out to every entitled recipient at once, pushed to platforms and alerts in the same moment. This is the research equivalent of an allocation decision — deciding who gets what — except that the answer is deliberately uniform: everyone entitled gets it, together.
"Entitled" is the crucial qualifier, and after unbundling it became genuinely complex. Not every client is entitled to every note. A client who pays for the industrials research gets the Meridian note; a client who does not is not entitled to see it. Before unbundling, research was sprayed widely because it was free; after unbundling, entitlements had to be enforced precisely, because a client seeing research they have not paid for is both a revenue leak and a compliance problem. So the publish step is really two mechanisms at once: fair and simultaneous distribution to everyone entitled, and rigorous enforcement of who is entitled in the first place.
Fair simultaneous distribution is a systems problem before it is a policy. The publishing platform holds the note under embargo, resolves the entitlement list — who has paid for this coverage, across every channel the client uses — and releases to all of them in one action, timestamped for audit. Get the timing wrong and one client's terminal shows the note seconds before another's inbox, and the bank has breached the fair-distribution rule without anyone intending to. Get the entitlements wrong and either a paying client is denied research or a non-paying one receives it. This is why the entitlements-and-distribution engine, unglamorous as it is, is one of the most heavily engineered parts of the research platform. Part 4 takes it apart.
Step 6 — Marketing turns a published note into flow, through two very different channels
Publishing the note is not the end of the analyst's job; it is the start of the selling. A note that publishes and is never marketed generates little. So research sales and the analyst go to work: they call key accounts, the analyst hosts a conference call to walk clients through the Meridian thesis, and she offers the highest-value service of all — access to the company's management. The written note is often the least valued part of the package. Clients frequently prize direct analyst access — the ability to interrogate the person behind the call — above the note itself, and corporate access above that.
The same Meridian view now travels down two channels with completely different rules. To institutional clients, it is general research: research sales markets the call, the fund decides for itself, and it trades directly through the bank's desks. No suitability duty applies, because the institution is presumed sophisticated. To wealth and private-banking clients, the view becomes the raw material for personal advice: an adviser translates the Meridian thesis into a recommendation tailored to a specific client's portfolio and objectives — and now the suitability duty bites. The insight the analyst created is the same; the obligations attached to delivering it could not be more different.
Corporate access is the crown jewel, and the most heavily policed service
The service the buy-side values most is not the note but the meeting: direct access to the companies it invests in. Through its analysts and relationships, a bank can arrange roadshows and one-on-ones with a company's chief executive or finance director, conferences that gather companies and investors, and field trips and expert calls for deeper insight. Surveys consistently rank corporate access among the highest-value services a bank provides — often above written research. It is also the hardest to allocate fairly, because management time is scarce and every client wants it.
Precisely because it is so valuable, corporate access is where the conflicts concentrate. Four risks run through it. It could become an inducement — "trade with us and you get access to the chief executive" — which is exactly the bundling MiFID II outlawed, so under European rules access must be priced and paid for separately, not handed out free with execution. Scarce management time must be allocated fairly, not simply auctioned to the biggest commission payers. Management meetings can stray into material non-public information, so they are tightly controlled under the Market Abuse Regulation. And who met whom, when, and what was discussed must be logged. Corporate access is a high-value service wrapped in a thick layer of controls, and managing it is now a distinct operational and technology workstream in its own right.
The advisory route carries a suitability duty the institutional route never triggers
When the Meridian view becomes a personal recommendation, the firm takes on responsibility for whether it fits the client — and must prove it. Suitability requires the adviser to know three things about the client before recommending anything: their knowledge and experience, so they can understand the product; their financial situation, so they can bear the risk; and their objectives and risk tolerance, so the recommendation serves what they are actually trying to achieve. The adviser gathers this in a suitability assessment, documents it, and must be able to produce a suitability report showing the advice fit. For packaged retail products, the client must also receive a standardised key information document under the PRIIPs rules. None of this applies when the same view is handed to an institution as general research.
Suitability is why advisory, especially private banking, is so process- and documentation-heavy. Each recommendation generates a paper trail — the assessment, the report, the target-market check confirming the product is being sold to the kind of client it was designed for. In a continuing relationship, suitability must be reviewed periodically, not just at the point of sale. This is a major systems and controls domain, and it is largest in Europe, where private-banking pools in France, Switzerland and Luxembourg make the advisory route a substantial business rather than an afterthought. Part 4 treats the systems that carry it.
Independent versus non-independent advice
When the Meridian view becomes personal advice, one more distinction bites: whether the advice is "independent." MiFID II separates independent advice, which must assess a broad range of the market and cannot retain inducements, from non-independent advice, which need not. The label is not cosmetic — it changes what the adviser must do and what they must tell the client. An adviser offering independent advice must be able to show they considered a wide field, not just the bank's own products; one offering non-independent advice must disclose that limitation up front. For a wealth client, knowing which kind they are receiving is central to knowing how much the recommendation has been shaped by the bank's own shelf. The suitability duty applies to both; independence governs the breadth of what was considered.
Step 7 — The value is captured only if consumption is measured, and the call is tracked for years
A note that is read but not measured cannot be charged for, so the last step closes the commercial loop. Before unbundling, this step barely existed — research was free, so nobody needed to know who read it. After unbundling, measuring consumption became business-critical, because you cannot price or defend research you cannot track. Every touchpoint is now recorded: who opened the Meridian note and for how long, who dialled into the analyst call, who requested a follow-up model, who used the corporate-access meeting. That consumption data feeds the annual negotiation in which each client's research fee is set from evidence rather than guesswork. This is the research equivalent of settlement: the moment the value created upstream is actually collected.
The analyst's job on Meridian does not end at publication; the call lives for as long as the rating stands. She updates the view as events unfold — each earnings release either confirms the margin inflection or challenges it — and the performance of the call is tracked against the target. If the thesis plays out and the shares approach 26, the track record strengthens and so does the franchise. If the thesis breaks, she must change the rating, publicly and through the same gate and the same fair-distribution rule as the original note. A rating is a standing commitment, not a one-time statement, and maintaining it is continuous work. Reputation — the analyst's real currency — is built and lost on this record over years.
The note is not one product but several, tuned to the moment
"The note" is a convenient singular, but a research desk produces several distinct kinds, each with a different job and rhythm. An initiation is the full statement of a new view — the fifty-page Meridian document that lays out the thesis, model, catalysts and risks from scratch. An update adjusts an existing view as facts change. An earnings flash is a fast reaction, sometimes minutes after a result, telling clients what just happened and whether it changes the call. A deep-dive is a long, considered piece of primary analysis on a single question. And a thematic note carries one big idea across many names or asset classes. The same analyst produces all of them, but the production rhythm and the client each serves differ sharply.
| Note type | What it does | Speed | Which client values it most |
|---|---|---|---|
| Initiation | Full first statement of a new view | Weeks | Long-only, pension (patient money) |
| Update | Adjusts an existing rating or target | Days | All clients holding the name |
| Earnings flash | Fast reaction to a result | Minutes to hours | Hedge funds (timely edge) |
| Deep-dive | Primary analysis of one question | Weeks | Long-only, thematic investors |
| Thematic | One idea across many names | Weeks | Cross-asset and macro clients |
Matching the note type to the moment and the client is part of the craft, and it is where automation now bites hardest. An earnings flash lives or dies on speed, which is exactly the mechanical work that AI-assisted summarisation compresses. An initiation lives on depth and differentiation, which stays stubbornly human. Knowing which kind of note a situation calls for — and which client will pay for it — is the difference between a desk that produces volume and one that produces value.
The analyst's second job is marketing, and it runs on a calendar
Producing the view is only half the analyst's role; the other half is selling it, and that work is relentless and scheduled. A lead analyst spends a large share of their time not writing but marketing — calls with key accounts, one-on-one meetings, conference presentations, and hosting the management access that clients prize. This is the research equivalent of a roadshow, except it never ends: the calendar of earnings seasons, conferences and corporate-access events sets a rhythm the analyst markets into continuously. A brilliant note that is never marketed generates little; a decent note marketed hard to the right accounts can move real flow.
The marketing effort is also where the wall is most easily tested, so it is tightly governed. When an analyst pre-markets a view or gauges client interest ahead of a big call, they operate close to the line that separates legitimate marketing from selective disclosure. The rule from Part 2 holds without exception: nothing material reaches one client before the fair, simultaneous publication reaches all. Pre-marketing discusses interest and themes, never an unpublished rating or target. The discipline is that the selling, however aggressive, never front-runs the publish gate.
At which stage does a research note become irreversible, and why does that matter?
At publication (stage five), after the compliance gate. A draft can be rewritten, but a published note is a fact in the market that cannot be recalled, which is why the review-and-publish moment carries the most compliance risk.
Why is a price target triangulated from multiple methods rather than taken from a discounted cash flow alone?
Because a DCF is highly sensitive to its assumptions; multiples anchor it to what the market pays for comparable companies, and the convergence of independent methods is what makes the target defensible.
What is a "variant view," and why is it the whole point of a rating?
A forecast that differs defensibly from consensus. A view that matches consensus adds nothing because the market already agrees; the value is in the defensible gap between the analyst's numbers and the collective view.
Explain the fair, simultaneous distribution rule and why "entitled" matters.
Every entitled client must receive a note at the same instant, so none gets a selective informational edge. "Entitled" means only clients who pay for that coverage receive it, which after unbundling had to be enforced precisely to prevent revenue leakage and compliance breaches.
What three things must an adviser establish before giving personal advice, and what document must result?
The client's knowledge and experience, financial situation, and objectives and risk tolerance; the adviser must produce a documented suitability assessment and a suitability report showing the advice fit.
Why is corporate access both the most valued and the most heavily policed research service?
The buy-side values direct management access above written research, but it risks becoming an inducement, must be allocated fairly, can stray into inside information, and must be logged — so under MiFID II it must be priced and paid for separately and tightly controlled.
Why did measuring research consumption become business-critical only after 2018?
Before unbundling research was free, so nobody needed to track readership; after unbundling, research must be priced and defended from evidence, and you cannot price or defend consumption you cannot measure.
What is the internal gauntlet a note clears at the compliance gate?
The restricted list (barring comment where the bank has MNPI or an active deal), quiet periods (around deals the bank is involved in), and conflict clearance (disclosures checked against the bank's actual relationships) — plus supervisory review of accuracy and balance.
2
When does a research note become irreversible?
It is irreversible from publication.
Right. A published note is a fact in the market.
They cannot be recalled.
Drafts can be rewritten.
Why triangulate a price target instead of using a DCF alone?
They have their own biases.
No.
They are widely used.
Right. Convergence of methods makes it credible.
What is a variant view?
Right. Agreeing with consensus adds nothing.
No.
No.
That is consensus.
What does “entitled” mean in the fair, simultaneous distribution rule?
Entitlements, not personal choice, decide.
Only entitled clients do.
Right.
That is exactly the breach.
3Roles, Actors, and Interactions
A note does not produce itself. It is the output of a small ecosystem — analysts who form the view, sales and advisers who carry it, clients who consume it, and companies who supply the access — with a compliance function threaded through every interaction. This part maps who is involved, who talks to whom, and why an information barrier has to cut straight through the middle of the same institution.
Four camps surround research, and a wall runs through the middle of them
The research ecosystem sorts into four camps, and the producing bank sits at the centre of almost every interaction. The first camp produces: analysts and strategists who turn data into views. The second distributes: research sales who carry views to institutions, and advisers who turn them into personal recommendations for wealth clients. The third consumes: the buy-side, from giant pension funds to individual private-banking clients. The fourth supplies and surrounds: the covered companies whose investor-relations teams provide access, and the independents and data vendors who both compete with the bank and sell to it. Across all four sits compliance, because independence and fair distribution are the product's foundation, not a side function.
What makes research different from a trading or banking business is that its central actor is deliberately walled off from its own colleagues. In most of a bank, the flow of information between desks is a feature. In research, it is a hazard. The analyst must be separated from the bankers who might want a favourable view and partly from the traders whose positions a rating could move. So the ecosystem is not a free flow of information around a hub; it is a set of carefully governed channels, some of them deliberately blocked. Mapping the actors means mapping both who talks to whom and who is forbidden from talking to whom.
| From | To | What flows | Stage |
|---|---|---|---|
| Covered company (IR) | Analyst | Filings, disclosures, management access | Analysis |
| Data & alt-data vendors | Analyst | Fundamentals, market data, alternative signals | Analysis |
| Analyst | Supervisory analyst & compliance | Draft note for review | Review |
| Publishing system | All entitled clients | The published note, simultaneously | Publication |
| Research sales | Institutional clients | The marketed view, analyst access | Distribution |
| Adviser | Wealth clients | A suitable personal recommendation | Distribution |
| Clients | Consumption analytics | Readership, interactions, meetings used | Value capture |
| Investment banking | Analyst | Blocked by the information wall | — |
The buy-side is not one client but five, and each consumes research differently
"The buy-side" is a convenient shorthand that hides five distinct kinds of investor, and research must be tailored to each. They differ in how long they hold, how sensitive they are to price, how they consume research, and what they will actually pay for. Getting this taxonomy right is what lets a research business target its scarce, expensive analyst time at the clients who value it most — the commercial heart of the post-unbundling world.
| Investor type | Holding horizon | How they consume research | What they pay for |
|---|---|---|---|
| Long-only funds | Months to years | The thesis and the analyst's judgement | Differentiated views, corporate access |
| Pension & sovereign wealth | Years to decades | Deep, considered research; management access | Long-horizon conviction, patient analysis |
| Hedge funds | Days to months | The call, the change, fast | Timely edge, bespoke work, expert access |
| Quant & systematic | Algorithmic, varies | Structured data, not prose | Machine-readable estimates and data feeds |
| Retail & wealth | Varies, adviser-led | Advice, via a private-banking adviser | Suitable, personalised recommendations |
Long-only funds and pension funds are the patient money, and they prize the analyst's judgement. A long-only fund holds for months or years and reads research for the thesis — the differentiated argument that a stock like Meridian is mispriced. Pension and sovereign wealth funds hold longer still, sometimes for decades, and value deep, considered research and direct management access above short-term calls. These are the clients an under-covered mid-cap initiation is aimed at: patient investors willing to back a conviction view and hold it while it plays out.
Hedge funds and quant funds are the opposite ends of a spectrum, and they stress the business in different ways. A hedge fund holds for days to months, wants the call and the change delivered fast, and pays for a timely edge, bespoke work, and expert access — but it also contributes the liquidity and trading volume the desks need. Quant and systematic funds barely read prose at all; they want the analyst's estimates as structured data they can feed into a model, which is why machine-readable output has become a real deliverable rather than a nicety. Retail and wealth clients form the fifth group, reached not directly but through an adviser, and consuming research only as it is translated into suitable personal advice.
Information barriers turn the org chart into an access-control problem
The defining control of the research business is the information barrier, and it is a genuine engineering problem, not a slogan. The barrier — the "Chinese wall" — separates research from the parts of the bank whose interests could corrupt its views. The problem it solves is peculiar: one institution knows things it must prevent itself from sharing across its own internal boundaries. The bankers on the private side know about deals in progress; the analysts on the public side must not. The trading desk knows its positions; the analysts must not be steered by them. The wall is the mechanism that keeps a single legal entity behaving as if it were several separate ones.
Four functions must be held apart, and the separations are not all identical. Research must be walled off firmly from investment banking, because the conflict there is direct and severe — a bank wanting a company's mandate has every incentive to want a favourable rating. Research must be separated from proprietary and desk trading, so that a rating is not shaped to help the book. Research must be insulated from material non-public information, so that it never publishes or trades on it. And the analyst's own personal dealing must be walled off from the analyst's professional view, through pre-clearance and disclosure. The IB separation is the hardest wall; the trading separation is partial, because research and sales-trading do legitimately interact to serve clients.
Wall-crossing is the controlled exception, not a breach
Sometimes the wall must be crossed on purpose, and the point is to do it in a logged, controlled way. Occasionally an analyst genuinely needs to be brought "over the wall" — for instance, to be told a company they cover is involved in a deal, so they can be restricted from publishing. This wall-crossing is not a failure of the barrier; it is its release valve. The crossing is formally requested, approved, logged, and time-bounded, and the analyst is added to a restricted list for the duration. The existence of a clean wall-crossing process is what lets the wall be absolute the rest of the time: exceptions are handled through the front door, under audit, rather than through quiet conversations.
The information wall is a technology system, not just a policy statement, and it lives in five interlocking controls. Surveillance monitors communications — email, chat, voice — for signs the wall has been breached. The restricted list, fed automatically from the deal pipeline, blocks research on any name where the bank has MNPI or an active transaction. Access controls physically prevent public-side systems from reaching private-side information. A wall-crossing log records every controlled exception with who, when, and why. And personal-account-dealing systems monitor analysts' own trades against their published views. A wall that exists only in a compliance manual is not a wall; the wall is these systems, running continuously.
Six internal teams run the research business, and the wall runs through them
The research organisation is a small number of specialised roles, and each maps to a stage and a system. Understanding who does what — and where the wall cuts through the org chart — is what lets an outsider navigate the business. Six roles cover it.
| Team | Core role | Primary technology |
|---|---|---|
| Lead analyst / strategist | Owns coverage, forms the view, sets ratings and targets, markets the call | Authoring & modelling platform |
| Research associate (RA) | Builds and maintains models, drafts sections, gathers data | Modelling tools, data feeds |
| Supervisory analyst / editor | Reviews every note for accuracy, balance and compliance | Review & workflow system |
| Research sales / specialist | Distributes and explains research to institutional clients | Research CRM, distribution platform |
| Research management / COO | Runs coverage strategy, budgets, productisation, the operating model | Consumption analytics, MIS |
| IT / Data / Change | Owns the platform: authoring, entitlements, distribution, analytics | The whole research technology stack |
The lead analyst is the revenue-facing expert and the face of the franchise; everyone else exists to leverage or control that person. The associate is the engine room — building models, gathering data, drafting — and the training ground for future analysts. The supervisory analyst is a required control, the human at the compliance gate. Research sales is the human channel to the institutional buy-side, translating and distributing the analyst's views. The wall runs cleanly here: none of these roles may be directed by investment banking, and their interactions with the private side are controlled and logged.
The two roles that grew most since unbundling are the ones furthest from the note itself. Research management and the chief operating officer own the survival agenda: which coverage pays for itself, how to productise and tier the offering, how to price analyst and corporate access, how to run the operating model efficiently. Before 2018 this was a light-touch function; today it is central, because someone has to answer the question unbundling forced — does this coverage earn its cost. And the IT, data and change function owns the platform on which the whole business now runs: authoring, entitlements, compliant distribution, and the consumption analytics that measure value. In a business where the routine is being automated and the value is being measured, the COO-and-technology axis is where transformation lands, and it is the natural home for change work.
Sell-side research now competes with a whole ecosystem, and buys from part of it
The sell-side bank is no longer the only source of investment insight, and the alternatives are both competitors and suppliers. Unbundling did not just shrink budgets; it opened the market to a whole ecosystem of providers that the buy-side can now pay directly. Understanding this ecosystem is essential, because it explains why "just another note provider" is a losing position and where a bank must be differentiated to survive.
| Provider | What they offer | Competitor, supplier, or both |
|---|---|---|
| Independent research providers | Conflict-free research sold directly | Competitor (helped by unbundling) |
| Expert networks | Paid calls with industry experts | Competitor for primary insight |
| Alternative-data vendors | Card, satellite, web, footfall data | Supplier and competitor |
| Data & platform vendors | Consensus, screening, analytics, terminals | Supplier; commoditises the data layer |
Independent research providers were the clearest winners from unbundling, because clients can now pay them explicitly. When research was bundled into commissions, an IRP had nothing to sell against "free" bank research; once clients paid separately, a conflict-free boutique competing on quality alone became a real alternative. Expert networks — firms that connect investors with industry specialists for paid calls — compete for the primary insight that used to flow only through the bank's analysts, and they are tightly compliance-managed because expert calls can stray into inside information.
Alternative-data and platform vendors sit on both sides of the line at once, which is the awkward part. The buy-side buys card-spending, satellite and web-scraped data directly, using it to get an edge before traditional research reflects it — competing with the bank. Yet the bank buys the same data to power its own research, and buys Bloomberg, FactSet and LSEG terminals that both distribute its notes and commoditise the consensus-and-screening layer research used to own. The strategic conclusion is sharp: a bank stays relevant only by offering what these alternatives cannot — deep sector expertise, differentiated analysis, direct analyst access and corporate access, delivered with data. Being a commodity note provider is a losing position, because the ecosystem now supplies commodity notes for less.
The interactions are governed channels, not a free flow
Because the wall runs through the ecosystem, information moves along defined channels with defined rules, and it helps to trace them concretely. A covered company's investor-relations team supplies filings and arranges management access to the analyst — a channel that must stay the right side of inside information. The analyst's draft flows only to the supervisory analyst and compliance, never sideways to the bankers. The published note flows to all entitled clients simultaneously, then to research sales who market it, and to advisers who translate it. Consumption data flows back from every channel to the analytics that price the research. And one channel is deliberately severed: investment banking cannot direct or preview the analyst's view. Reading the ecosystem as a set of permitted and forbidden channels — rather than a free flow around a hub — is what makes the wall comprehensible as an operating model rather than a slogan.
Why does an information barrier have to run through the middle of a single institution?
Because one legal entity knows things it must prevent itself from sharing internally — bankers know about deals and traders know their positions that analysts must not — so the wall keeps one institution behaving as if it were several separate ones.
Name the five buy-side investor types and one way each consumes research differently.
Long-only (reads the thesis, long horizon), pension/SWF (deep research, longest horizon), hedge funds (wants the call fast), quant/systematic (wants structured data not prose), and retail/wealth (consumes research only as suitable advice via an adviser).
How does the buy-side taxonomy change how a research business is run?
It lets the business target scarce analyst time using consumption data — deep initiations aimed at long-only and pension money, fast reactions at hedge funds, clean data feeds at quants — turning a fixed budget into a targeted one.
What is wall-crossing, and why is it a control rather than a breach?
The formal, logged, time-bounded process of bringing an analyst over the wall when they genuinely need private-side information; it is the wall's controlled release valve, and having a clean process is what lets the wall stay absolute the rest of the time.
Why is the information wall described as a technology problem, not just a policy?
Because it lives as running systems — communications surveillance, an automated restricted list, access controls, wall-crossing logs and personal-dealing monitoring — and the expanding surface of new messaging channels means it must be continually re-secured, not built once.
Which two research roles grew most after unbundling, and why?
Research management/COO and the IT/data/change function, because unbundling forced the business to prove which coverage earns its cost, to productise and price its services, and to run on a platform that measures consumption — all owned by the COO-and-technology axis.
3
Why must an information barrier run through the middle of one institution?
No.
Right.
No.
No.
What does a quant/systematic investor want from research?
Right.
No.
That suits long-only investors.
That is the hedge fund's priority.
Why is wall-crossing a control rather than a breach?
Right. A clean log is evidence the wall works.
Frequency is not the test.
It is logged, not secret.
A crossed analyst is restricted.
How does the buy-side taxonomy change how research is run?
Right.
No.
It directs where effort goes.
Ratings must be independent of clients.
4Systems, Data, and IT
A research business used to run on models in spreadsheets and notes in email. It cannot any longer. Post-unbundling, the platform that authors, reviews, publishes, entitles, distributes and measures research is as much the business as the analysts are. This part builds that platform from the clean question — what functions does any research operation need — before naming a single vendor, then follows one note through the systems, takes apart the data model, and reads the five regulations that shape every box on the diagram.
The functional architecture is a chain of six blocks with a compliance spine
Before asking which vendor does what, ask what functions a research operation cannot do without. Strip away the brand names and every research business needs the same six capabilities, in sequence, from the analyst's desk to the client's screen and back. Getting this functional view clear first is what stops a technology discussion collapsing into a list of products.
| Functional block | What it does | Stage it serves |
|---|---|---|
| Authoring & modelling | Build models, draft notes, link to data | Analysis, drafting |
| Review & approval workflow | Route drafts through supervisory and compliance sign-off | The compliance gate |
| Publishing & entitlements | Publish once; control who is entitled to see it | Publication |
| Distribution | Push to terminals, portals, apps, email, aggregators | Distribution |
| Consumption analytics & CRM | Track readership, interactions, events; link to clients | Value capture |
| Data foundation | Source, clean, integrate and govern all input data | Underlies everything |
The six blocks form a chain, but a compliance-and-entitlements spine runs through all of them at once. Data feeds authoring at the front. Consumption data flows back to the business at the end. And threaded through the whole chain is the cross-cutting layer that makes research research: the information-wall controls, the entitlement rules that decide who sees what, the disclosure engine, the reference data, and the audit trail. That spine is why a research platform cannot be assembled from generic publishing tools — the compliance and entitlement logic is not a feature you add, it is the structure the whole thing hangs on.
Where the chain breaks in practice
The most common failure in a real research operation is not a missing block but blocks that do not join up. Authoring, distribution and the client CRM are frequently separate systems that were bought at different times and never integrated, so no one can see a single view of a client or of the value the business delivers. Entitlements — who can see what, post-unbundling — are intricate and error-prone when spread across disconnected platforms. Consumption has blind spots wherever a distribution channel is not instrumented, and if you cannot see usage you cannot price it. And compliance overhead balloons when disclosures, distribution rules and record-keeping are done by hand. The single most valuable thing a transformation can do to a research platform is join the chain end to end.
Real vendors fill each block, and where a bank builds tells you what it thinks is proprietary
The functional blocks map onto a recognisable vendor landscape, and the buy-versus-build pattern is a map of competitive advantage. Banks buy the commoditised plumbing and build only what differentiates them. Knowing which is which tells you where the business thinks its edge lives.
| Block | Typical vendors / approach | Buy or build |
|---|---|---|
| Research management system (RMS) | Specialist platforms (e.g. Bluematrix and similar) | Buy |
| Data & distribution terminals | Bloomberg, FactSet, LSEG (Refinitiv) | Buy |
| Aggregators & marketplaces | Research discovery and payment platforms | Buy |
| Consumption analytics / research CRM | Specialist analytics vendors or built in-house | Buy or build |
| Data & alt-data | Fundamentals, market data, alternative-data vendors | Buy |
| AI / LLM tooling | Foundation models plus a governed internal layer | Buy the model, build the controls |
The research management system is the production backbone, and almost every bank buys it. An RMS authors, formats, controls, publishes and entitles research across channels — it is the spine of the six-block chain rendered as a single product. Banks buy it because it is a solved problem: the workflow of drafting, reviewing, disclosing and publishing is common across the industry, so building it in-house would burn money reinventing a commodity. The terminals — Bloomberg, FactSet, LSEG — are where clients actually read research and pull data, and they are bought for the same reason: no bank can out-build Bloomberg's distribution reach.
What banks build, or heavily customise, is the part that measures and prices value — because that is where the edge now lives. Consumption analytics and research CRM sit on the boundary: some banks buy specialist tools, others build, because the exact way a bank measures and attributes research value is increasingly a source of competitive advantage in the post-unbundling fight. And the decision engine that turns consumption into a client price is almost always proprietary, because it encodes the bank's own commercial judgement about what its research is worth. The pattern holds across the whole of capital markets: banks buy the workflow and build the intelligence.
One note's journey shows how a data error cascades through every downstream system
Follow a single object — the Meridian note and its rating — from creation to the client fee, and the architecture stops being abstract. Each system in the chain consumes what the previous one produced, which means a mistake at the top reaches everything below it. This is the cascade principle, and it is the reason data quality in research is not a hygiene issue but a business risk.
| Stage | System | Data object created |
|---|---|---|
| Analysis | Authoring & modelling | The model and the rating (Buy, target 26) |
| Review | Workflow & disclosure engine | The approved note plus its disclosures |
| Publication | Publishing & entitlements | The published note, tagged to an entitlement list |
| Distribution | Distribution platform | Delivery events across every channel |
| Consumption | Analytics & CRM | Readership and interaction records per client |
| Value capture | Pricing / decision engine | The client's research fee |
An error at the top of the chain is not contained; it is amplified. Suppose the rating is tagged to the wrong entitlement list at publication. The distribution platform faithfully sends the Meridian note to clients who have not paid for it — a revenue leak and a compliance breach at once. The consumption analytics then record readership for the wrong clients, so the pricing engine sets fees from corrupted evidence. One mis-tag at the source becomes a distribution failure, a compliance incident, and a mispriced client, because every downstream system trusted the upstream data. The cascade is why research platforms invest so heavily in getting the object right at the point of creation.
The atom of the business is a structured record, not a PDF
The published note looks like a document, but the system treats it as a structured object with fields that drive everything downstream. A PDF is what the client reads; the record behind it is what the platform acts on. That record carries the security identifier, the rating, the price target, the horizon, the analyst, the publication timestamp, the disclosure set, and the entitlement tag. Each field does work. The identifier links the note to the right instrument. The rating and target feed alerts and track records. The timestamp proves fair simultaneous distribution. The entitlement tag decides who receives it. Treating research as data rather than documents is precisely what lets it be distributed to machines, tracked at scale, and priced.
The hardest data problem in research is not the note — it is the client. A single buy-side firm reaches the bank through many identities: a Bloomberg address, a FactSet login, an email domain, a corporate-access booking, a phone call logged in the CRM. To price research per client, all of those have to be resolved to one entity, and that deduplication is genuinely difficult — the same fund appears under slightly different names on every channel. Get it wrong and consumption is double-counted or lost, and the fee is set from a broken picture. Client-identity resolution is the unglamorous, mission-critical plumbing beneath every consumption analytic, and it is where a great deal of research-technology effort quietly goes.
Identifiers keep the chain honest
Three identifiers do the work of tying a note to the right instrument, entity and structure, and each bites somewhere different. The ISIN identifies the security — the specific share of Meridian the note covers. The LEI identifies the legal entity — the fund receiving and paying for the research, and the company being covered. And instrument-mapping identifiers, together with vendor tickers and codes, let the same security be recognised across Bloomberg, FactSet and LSEG, which each use their own symbology. Without clean identifiers the chain silently misroutes: a note tagged to the wrong ISIN reaches the wrong watchlists, and consumption booked against the wrong LEI is billed to the wrong client.
| Identifier | Identifies | Where it bites |
|---|---|---|
| ISIN | The security covered | Linking the note to the right instrument and alerts |
| LEI | The legal entity (client or covered company) | Billing consumption to the right client |
| Ticker / RIC / vendor code | The security across data vendors | Recognising the same instrument across terminals |
The note object, field by field, and why each field is load-bearing
If research is data rather than documents, then the fields of the note object are the real product specification, and each one drives a downstream system. A useful way to see the platform is to open up the record behind the Meridian note and ask, for every field, what breaks if it is wrong. The answer is never "nothing." Each field is consumed by a system that trusts it, so the field list is also a map of dependencies.
| Field | What it holds | What it drives downstream |
|---|---|---|
| Note ID | Unique identifier for this publication | Audit trail, versioning, consumption linkage |
| ISIN / instrument | The security covered | Alerts, watchlists, portfolio matching |
| Rating | Buy / Hold / Sell | Track record, rating-distribution disclosure, alerts |
| Price target & horizon | The level and its timeframe | Performance measurement, client screens |
| Analyst & desk | Author and coverage team | Reputation records, compensation, conflict checks |
| Publication timestamp | The exact release moment | Proof of fair, simultaneous distribution |
| Disclosure set | Conflicts, relationships, methodology | Regulatory compliance, MAR fair presentation |
| Entitlement tag | Which clients may receive it | Distribution routing, revenue attribution |
| RIXML metadata | Machine-readable tags for all of the above | Structured delivery to quant clients and AI systems |
Two fields carry more weight than the rest, and they are the ones a transformation must never get wrong. The publication timestamp is the evidence that fair distribution happened — remove it, or record it inconsistently across channels, and the bank cannot prove it did not tip a favoured client. The entitlement tag decides both who legally receives the note and who is billed for it, so an error here is simultaneously a compliance breach and a revenue leak. The other fields matter, but these two are where a single bad value becomes a regulatory incident rather than an inconvenience.
The note object is versioned, not overwritten, and that design choice is a compliance requirement disguised as a data-modelling decision. When an analyst changes the Meridian rating from Buy to Hold, the platform does not edit the old record — it creates a new version and preserves the old one immutably. The reason is that a regulator, or a client dispute, may need to reconstruct exactly what the bank said and when. An append-only version history means the answer to "what was your rating on this date" is always recoverable and always tamper-evident. Mutable records save storage; immutable versioned records save the bank when the question is asked years later.
Research travels in two data worlds: structured content and the channels that carry it
Just as a capital markets deal splits into a fast front-office messaging world and a slower post-trade one, research splits into structured content and distribution transport. The content layer is the machine-readable representation of the note itself — increasingly delivered as structured data, not only as a PDF, so that a quant client's systems can ingest the rating, the estimates and the target directly. The industry standard for this is RIXML, a markup that tags a research note's metadata and content so any system can parse who wrote it, what it covers, the rating, and the disclosures. The transport layer is the set of channels that carry the content: terminals, client portals, mobile apps, email, and third-party aggregators, each with its own delivery and entitlement mechanics.
| Layer | Standard / channel | Carries | Timescale |
|---|---|---|---|
| Content | RIXML structured research | Note metadata, rating, estimates, disclosures | Per publication |
| Distribution | Terminals, portals, apps, email, aggregators | The note to entitled clients | Simultaneous release |
| Consumption | Analytics feeds from every channel | Readership and interaction events | Continuous |
Integration patterns decide how fresh the data is
The three ways systems talk to each other — batch, API, and streaming — are the same across the bank, and research is migrating along the same path. Batch is the legacy default: data moves in overnight bundles, which is simple but stale, so a consumption report built on batch is always a day behind. An application programming interface lets a system pull data on demand, fresher but still request-driven. Event streaming pushes data the instant it changes — a note is published, and every downstream system knows immediately — which is what makes real-time consumption tracking and instant client alerts possible. The migration from batch toward streaming is not merely a speed upgrade; it removes whole categories of reconciliation, because systems that are pushed the same event in real time do not drift out of step and then have to be reconciled back together.
Immutable audit and the pricing engine are the two records that cannot be casual
Two kinds of record in research must be built to a higher standard than the rest. Compliance-critical records — who approved a note, when it published, who was entitled, who received it — must be append-only and immutable, because they are the evidence the bank produces when a regulator asks whether distribution was fair and disclosures complete. You cannot let anyone quietly edit the history of who received a note; the audit trail has to be tamper-evident by construction. And the pricing-and-decision engine that turns consumption into a client fee is almost always proprietary, because it encodes the bank's own commercial judgement — the same reason an allocation engine in capital markets is never bought off the shelf.
Five regulations shape every box on the diagram, and each reads as a system specification
A research platform is not designed against a blank sheet; it is designed against five regulatory regimes, and each one dictates concrete system behaviour. Read as system specifications rather than legal summaries, they explain why the architecture looks the way it does.
| Regulation | What it governs | System it dictates |
|---|---|---|
| MiFID II (unbundling, inducements, independence) | Separate payment for research; research independence | Entitlements, payment plumbing, RPAs, the wall |
| MAR (Market Abuse Regulation) | Fair presentation of recommendations; inside information | Disclosure engine, restricted lists, surveillance |
| Suitability & product governance (incl. PRIIPs) | Personal advice must fit the client | Suitability assessment tools, KID generation |
| GDPR | Personal data, including in alternative data | Data governance, consent, retention controls |
| EU AI Act | Governance of AI systems | Model-risk controls, human-in-the-loop, logging |
MiFID II and MAR shape the core of the platform. MiFID II's unbundling rule is why the entitlements-and-payment plumbing exists at all — the system that tracks who is entitled to what and who has paid is a direct response to the requirement to price research separately, and it must now flex to accommodate the Listing Act's partial re-bundling. Its independence provisions are why the information wall is engineered as access controls and surveillance. MAR requires investment recommendations to be fairly presented, with sources, methodology and interests disclosed, which is why the disclosure engine and the restricted list are load-bearing, and it governs inside information, which is why communications surveillance runs continuously.
The other three regimes shape the edges that have grown fastest. Suitability and product governance dictate the assessment tools and the key-information-document generation that make the advisory route so documentation-heavy — strongest for retail and wealth clients, and a major European systems domain. GDPR bites hardest on alternative data: card-spending and web data are often personal data, so sourcing and using them requires consent, minimisation and retention controls, a real constraint on the data foundation. And the EU AI Act, the newest of the five, governs how AI systems are deployed — pushing banks toward controlled, logged, human-in-the-loop use of large language models in research rather than autonomous generation. Each regulation is, in effect, a requirements document for a part of the platform.
The data foundation is six layers deep, and it is the real competitive edge
Everything above rests on the data foundation, and in a world where clients can buy raw data themselves, how a bank combines it is the edge. Research has always been data-driven, but the data layer used to be a shared utility — everyone read the same filings. Today the buy-side can buy fundamentals, market data, even alternative data directly, so the analyst's advantage comes not from having data but from combining superior data with judgement no one else has. Six layers make up the modern research data stack, and building, cleaning and governing them is the unglamorous prerequisite for both better research and any AI on top of it.
| Data layer | What it is | What it powers |
|---|---|---|
| Fundamentals & filings | Financials, filings, transcripts | The bedrock of equity and credit analysis |
| Market & pricing data | Prices, spreads, volatilities, flows | The market context for every call |
| Estimates & consensus | The analyst's and the market's forecasts | Finding the variant view |
| Alternative data | Card, satellite, web, footfall, shipping | Early signals before fundamentals report |
| ESG & sustainability | Emissions, taxonomy, governance scores | A required input and a growth theme |
| Internal & interaction | The bank's own flow and consumption data | Targeting research and proving value |
The estimates layer is where the variant view is found, and the alt-data layer is where the new edge is fought for. Consensus — the aggregated forecasts of everyone covering a name — is the reference point against which the analyst locates her differentiated view; without a clean consensus feed, she cannot know where she disagrees with the market. Alternative data is the frontier: card-spending or satellite imagery can hint at a company's performance before it reports, giving the analyst a signal ahead of the crowd. But most alt-data contains no durable edge, it is expensive, and its licence terms on redistribution inside published research are genuinely tricky. The honest truth of the data stack is that the real work is plumbing — sourcing, cleaning, integrating and governing — and that plumbing is the prerequisite for everything glamorous built on top.
Consumption analytics is the system that turned research into a measured business
Of all the systems in the platform, the one unbundling made indispensable is the one that measures who consumes research. Before 2018 nobody needed it; after 2018 you cannot price or defend research you cannot measure, so consumption analytics went from a nice-to-have to the beating heart of the commercial operation. It tracks five things, and each feeds a decision the business could not otherwise make.
| What is tracked | Why it is tracked |
|---|---|
| Readership | Who opened which note, for how long, on which channel |
| Interactions | Analyst calls, meetings, bespoke requests — the high-value touchpoints |
| Events & access | Conference attendance and corporate-access meetings used |
| Entitlements | Who is contractually allowed to consume what |
| Payments | What each client pays, against what they consume |
The data does four jobs, and together they are the survival toolkit of a post-unbundling research business. It sets the price, letting the bank negotiate each client's fee from evidenced consumption rather than guesswork. It targets the spend, directing scarce analyst time toward the clients and topics that generate real value. It defends the budget, showing management and the client the value research delivers. And it keeps the business compliant, enforcing entitlements and fair distribution and evidencing that access was not an improper inducement. A research desk that cannot see its own consumption is flying blind on price, on strategy, and on compliance at once — which is why standing up this capability is one of the highest-value things a transformation can deliver.
Operational resilience: the publish moment has no second chance
Most of a research platform can tolerate a brief outage, but one moment cannot, and resilience is designed around it. When a market-moving note publishes, it must reach every entitled client at the same instant. If the distribution system fails halfway through the release, some clients have the note and others do not — and the bank has breached the fair-distribution rule not by intent but by infrastructure. The embargo-and-release mechanism is the research equivalent of a settlement window in capital markets: a narrow, irreversible moment where the systems simply cannot fail. Resilience engineering exists to guarantee that they do not.
| Concept | What it guarantees | Why it matters at the publish moment |
|---|---|---|
| Redundancy | A spare for every critical component | No single failure can halt a release |
| Automatic failover | Instant switch to the backup | The release completes even if primary fails |
| Disaster recovery | A tested plan to restore from a major outage | The business survives a data-centre loss |
| Business continuity | People and process keep running | Publication proceeds during disruption |
Redundancy alone is not enough, and that distinction is the heart of resilience. Having a backup system does nothing if switching to it is slow or manual, because the fair-distribution window is measured in seconds. What matters is automatic failover — the switch happening instantly and invisibly, so the simultaneous release completes even as the primary system dies. Disaster recovery and business continuity extend the same logic to larger shocks: a tested plan to restore from a data-centre loss, and the people and processes to keep publishing through disruption. Resilience is where "mission-critical" stops being a slogan and becomes a measurable guarantee about a specific irreversible moment.
The entitlements and payments engine is the hardest machine unbundling built
If one system defines the modern research platform, it is the entitlements engine, because unbundling turned "who can see this" from an afterthought into a precise, billable, auditable question. Before 2018 research was sprayed widely and entitlements barely existed. Today the platform must know, for every note and every client, whether that client is contractually entitled to receive it, across every channel the client uses, reconciled against what they have paid. This is genuinely hard, because entitlements are multi-dimensional and constantly changing, and because the same content flows through channels the bank does not fully control.
| Entitlement dimension | The question it answers | Why it is hard |
|---|---|---|
| Content scope | Which coverage, desks or products the client has bought | Packages and tiers change per client, per year |
| Channel | Terminal, portal, app, email, aggregator | Each channel enforces entitlements differently |
| Client identity | Which legal entities and users are covered | One firm has many identities to resolve |
| Time | When entitlement starts and ends | Trials, lapses and renewals must be exact |
| Payment status | Whether consumption is matched to payment | RPA reconciliation is a separate, fiddly ledger |
The payments half of the engine is a reconciliation problem, and it is where the Research Payment Account plumbing lives. When a client funds research through a ring-fenced account rather than the bank's own profit and loss, every euro of consumption has to be matched against the agreed budget and drawn down correctly. That is a ledger the bank did not need before unbundling and now cannot operate without — and the Listing Act's partial re-bundling adds yet another payment mode to reconcile alongside it. The engine that used to be a simple access list is now an entitlements-and-billing system with a compliance obligation attached to every row.
Entitlements fail in two opposite and equally damaging ways, which is why the engine is engineered so defensively. Over-entitlement sends a note to a client who has not paid — a revenue leak and, worse, a distribution that may look like an improper inducement. Under-entitlement denies a paying client research they are owed — a service failure that shows up in the next renewal negotiation. Between these sits the aggregator problem: once a note reaches a third-party research marketplace, the bank is relying on that platform to enforce entitlements it can no longer see directly. Getting entitlements right is therefore not one control but a chain of them, running from the bank's own systems out to platforms it does not own — which is exactly why it is the most heavily engineered and most frequently rebuilt part of the stack.
The research management system does five jobs behind one interface
The research management system is bought, not built, but understanding what it actually does explains why it is the spine of the platform. To an analyst it looks like a place to write a note. Underneath, it is performing five distinct jobs at once, and it is the integration of those jobs into one controlled workflow that makes it valuable — and hard to replace.
| Job the RMS does | What it means in practice |
|---|---|
| Authoring | A structured editor linking the note to models and live data |
| Formatting | House style, templates and consistent branding applied automatically |
| Disclosure automation | Assembling the correct conflict and relationship disclosures per note |
| Workflow & control | Routing the draft through supervisory and compliance sign-off |
| Publishing & entitling | Releasing once, tagged with entitlements, across every channel |
Disclosure automation is the job that quietly justifies the whole system, because doing it by hand is where breaches are born. Every note must carry the exact disclosures that reflect the bank's relationships with the covered company, the analyst's holdings, and the distribution of the bank's ratings. Assembling those manually invites the omission that becomes a MAR breach. The RMS pulls them from the bank's own records automatically, so the disclosure set is correct by construction rather than by the analyst's memory. Combined with the workflow that will not let a note publish without sign-off, the RMS turns the compliance gate from a human promise into a system guarantee — which is precisely why banks buy a specialist platform rather than assembling one from generic tools.
The research corpus is a searchable knowledge base, not a folder of PDFs
A bank publishes tens of thousands of notes a year, and the accumulated corpus is an asset only if it can be found, which makes search and metadata a first-class system. A decade of research is worthless if a client or an analyst cannot locate the relevant piece in seconds. So the modern platform treats the corpus not as a document archive but as a structured, indexed knowledge base, where every note is tagged with machine-readable metadata — the RIXML fields, the securities and sectors covered, the analyst, the date, the rating — and made searchable across all of them. The difference between a shared drive of PDFs and a governed research corpus is the difference between an archive and a product.
Two kinds of search now sit side by side, and the second is what makes AI over the corpus possible. Traditional keyword-and-metadata search answers precise questions: every note on Meridian, every downgrade in European industrials this quarter. Semantic search — powered by turning notes into numerical representations, or embeddings, stored in a vector index — answers questions of meaning: find research making an argument similar to this one, even if it uses different words. Semantic search is not a nicety; it is the retrieval layer that a large language model needs to answer questions across the corpus accurately, which is why building a well-tagged, well-indexed research library is the unglamorous prerequisite for any credible AI assistant on top of it.
The research corpus is where information security and knowledge management collide, because the same library that clients search also contains material non-public information and embargoed drafts. A search system over the corpus cannot simply return the most relevant note — it must return the most relevant note the searcher is entitled and cleared to see. That means entitlements and information-wall permissions have to be enforced inside the search layer itself, not bolted on afterward. A semantic search that ignores the wall would happily surface a restricted draft to someone who must not see it. Permission-aware retrieval — filtering results by entitlement and wall status before they are ever shown — is the control that makes a searchable corpus safe, and it is the same control an AI assistant must inherit.
The AI research assistant is a retrieval system with a compliance boundary
The credible way to put a large language model into research is not to let it write freely, but to wrap it in retrieval and controls — an architecture worth understanding because it is where most research-AI effort now goes. A raw model that generates from its training data alone will hallucinate figures and cannot cite a source, which is disqualifying for published research. The pattern that works is retrieval-augmented generation: the model is given the relevant, entitled documents from the corpus first, and asked to answer only from them, with citations. The analyst's judgement, the supervisory review, and fair distribution all remain exactly where they were. AI changes how the draft is produced, not who is accountable for it or how it is released.
Five design constraints separate a usable research assistant from a liability, and each maps to a risk from Part 5. The assistant must retrieve before it generates, so its output is grounded in real documents rather than invented. It must respect the information wall at the retrieval layer, so it never surfaces MNPI or restricted drafts to someone who must not see them. It must cite its sources, so the analyst can verify every claim. It must log every prompt and response, so the behaviour is auditable under model-risk governance and the EU AI Act. And it must keep the human in the loop as a hard gate, because accountability for a published note cannot be delegated to a model. Strip any one of these out and the tool moves from force-multiplier to incident waiting to happen.
| Design constraint | The risk it controls |
|---|---|
| Retrieve before generate | Hallucination and unsupported figures |
| Wall-aware retrieval | MNPI leakage and information-barrier breach |
| Cite every source | Unverifiable claims in published research |
| Log prompts and outputs | Model-risk governance, EU AI Act auditability |
| Human gate before publish | Loss of accountability for the view |
Model-risk governance treats an AI assistant the way the bank treats any model that touches a client outcome: it must be validated before use, monitored in production, explainable on demand, and owned by a named person. For a research assistant this means version-controlling the model and its prompts, testing it for accuracy and bias, and being able to reconstruct why it produced a given output. The governance is not a brake bolted on at the end; it is the difference between a tool a regulator will accept and one that becomes a finding. In Europe, the EU AI Act formalises much of this, pushing banks toward controlled, logged, human-in-the-loop deployment by design rather than by choice.
Cloud, APIs and the connection to the rest of the bank
Research used to run on isolated systems; modernisation is about turning the platform into a set of API-connected services that join the bank together. The historical research stack was a set of separate applications — an authoring tool here, a distribution system there, a CRM somewhere else — stitched loosely with overnight files. Cloud and API-first architecture rebuilds this as services that call each other on demand: the authoring system pulls live data through an API, the entitlements service answers "can this client see this" in real time, and consumption events stream back the instant they happen. This is what finally joins the six-block chain end to end, and it is the precondition for both structured client delivery and any AI on top.
Research does not live in isolation, and the integration points to the wider bank are where it earns its keep. The platform must connect to market-data feeds that price every security in every model, to the sales-trading systems where the flow research drives actually executes, to the client CRM that records every interaction, and to the compliance systems that feed the restricted list and run surveillance. Each integration is a place where research either plugs into the bank's nervous system or remains a silo. The strategic value of API-first architecture is not elegance; it is that a well-connected research platform can prove its contribution — linking a published view to the consumption it drove and the flow that resulted — which is exactly the attribution unbundling demands.
Information security: the corpus is a vault of inside information
A research platform holds some of the most sensitive information in the bank — unpublished ratings, embargoed drafts, and the material non-public information analysts are cleared into — which makes information security a core design concern, not an IT afterthought. An unpublished Meridian downgrade is price-sensitive: anyone who obtained it early could trade on it illegally. So the platform must protect the corpus the way a trading system protects positions, through layered controls that assume both external attackers and internal misuse.
| Control | What it protects against |
|---|---|
| Access control & least privilege | Staff seeing research or drafts they are not cleared for |
| Data-loss prevention (DLP) | Sensitive drafts leaving the bank by email or upload |
| Communications surveillance | The wall being breached over chat, email or voice |
| Encryption in transit and at rest | Interception of notes and the corpus |
| Immutable audit logging | Undetected tampering with what was published when |
The hardest information-security problem in research is the same as its hardest compliance problem: the enemy is often on the inside. Most controls defend against outsiders, but the research-specific risk is an insider — an analyst tipping a favoured client, a banker reaching across the wall, someone trading on an unpublished rating. That is why data-loss prevention and communications surveillance sit at the centre of the research security stack: they watch the legitimate channels for illegitimate use. Information security and the information wall are, in the end, the same discipline viewed from two angles — one protects the corpus from theft, the other protects it from misuse, and both live as running systems rather than policies on a page.
The advisory technology stack is a separate machine built around suitability
Everything so far describes the research platform; the advisory business runs on a different stack, built around the one obligation research never carries — proving each recommendation suits the client. When the Meridian view travels to a wealth client, it enters a world of systems research does not need: a suitability engine, portfolio tools, product-governance controls and document generation. This is a major European systems domain precisely because private banking is a large business in France, Switzerland and Luxembourg, and because the suitability duty makes every recommendation a documented, defensible event.
| Advisory system | What it does | The duty it serves |
|---|---|---|
| Suitability engine | Assesses client knowledge, situation, objectives, risk | The suitability obligation |
| Client profiling & KYC | Captures and maintains the client's profile | Know-your-client, appropriateness |
| Portfolio construction | Builds and rebalances portfolios and model portfolios | Consistent, scalable advice |
| Product governance | Enforces the target market for each product | Selling within the intended market |
| Document generation | Produces suitability reports and PRIIPs KIDs | Evidence and disclosure |
The suitability engine is the heart of the advisory stack, and it is what makes tailored advice scalable rather than artisanal. For each client it holds the profile — knowledge and experience, financial situation, objectives and risk tolerance — and checks every proposed recommendation against it before an adviser can act, generating the suitability report that evidences the fit. Model portfolios let a house view like the Meridian thesis be translated into consistent, suitable recommendations across many clients at once, rather than reconstructed by hand for each. And where advice becomes fully digital — hybrid and robo-advisory models that combine algorithms with human oversight — the same suitability logic runs automatically, scaling personalised, compliant advice to clients a human adviser could never reach one by one.
What are the six functional blocks of a research platform, and what runs through all of them?
Authoring & modelling, review & approval, publishing & entitlements, distribution, consumption analytics, and the data foundation — with a compliance-and-entitlements spine running through every block.
Why do banks buy the RMS and terminals but build or customise the consumption-and-pricing layer?
The production workflow and distribution reach are solved commodities not worth reinventing, while the way a bank measures and prices research value is a source of competitive advantage, so the decision engine is kept proprietary.
Explain the cascade principle using a mis-tagged entitlement. — If a note is tagged to the wrong entitlement list, distribution sends it to non-paying clients (a leak and a breach), analytics record the wrong readership, and the pricing engine sets fees from corrupted data — one upstream error is amplified through every downstream system.
Why is client-identity resolution described as mission-critical plumbing?
A single client reaches the bank through many identities across channels; unless they are deduplicated to one entity, consumption is double-counted or lost and the research fee is set from a broken picture.
Why does the move from PDF to structured (RIXML) research matter commercially?
Quant and systematic clients want machine-readable estimates and data, not prose; a bank that cannot deliver structured output is invisible to a growing part of the buy-side, so structuring output decides whether a whole client category can consume the product.
Why is migrating from batch to event streaming more than a speed upgrade?
Real-time push keeps downstream systems in step as events happen, which removes whole categories of reconciliation rather than simply delivering the same stale data faster.
Pick two of the five regulations and state the system each one dictates.
MiFID II dictates the entitlements-and-payment plumbing and the information wall; MAR dictates the disclosure engine, restricted lists and surveillance; suitability/PRIIPs dictates assessment tools and KID generation; GDPR dictates data-governance controls over alt-data; the EU AI Act dictates model-risk controls and human-in-the-loop logging.
Which two fields of the note object become a regulatory incident if wrong, and why?
The publication timestamp (the evidence of fair, simultaneous distribution) and the entitlement tag (which decides who legally receives the note and who is billed), so an error in either is a compliance breach rather than an inconvenience.
Why is the entitlements-and-payments engine the hardest machine unbundling built?
Entitlements are multi-dimensional (content, channel, identity, time, payment) and constantly changing, must be reconciled against RPA budgets, and must be enforced even out to aggregators the bank does not control; failure means either a revenue leak or a service failure.
What is retrieval-augmented generation, and why is it the credible way to use AI in research?
The model is given relevant, entitled documents from the corpus and answers only from them with citations, rather than generating freely; this grounds output in real sources, controls hallucination, and keeps the wall, logging and human gate intact.
Why must search over the research corpus be permission-aware?
The corpus contains MNPI and embargoed drafts, so search must return only the most relevant note the searcher is entitled and cleared to see; entitlements and wall status must be enforced inside the retrieval layer, and an AI assistant inherits the same control.
How does the advisory technology stack differ from the research platform, and why?
It is built around the suitability duty research never carries — a suitability engine, portfolio and model-portfolio tools, product governance and KID/report generation — because every personal recommendation must be shown to fit the client.
4
A note is tagged to the wrong entitlement list. What happens?
Right. The cascade principle.
No automatic recall.
They drive distribution.
No.
What do banks build or customise in a research platform?
Bought.
Bought.
Usually bought.
Right. Measuring and pricing research is an edge.
Why is client-identity resolution mission-critical?
Not the point.
Right.
It is.
No.
What runs through all six functional blocks of a research platform?
No.
No.
Right.
No.
5Risks, Pain Points, and Transformation
Research fails in a small number of characteristic ways, and each has a real case attached to it. This part maps where the risk lives, how to tell a good research business from a lucky one, and the three failures — a conflict scandal, a distribution breach, and a regulatory shock — that between them define the modern rulebook. Then it turns forward, to the five technologies genuinely changing the business and the honest place of generative AI within them.
A MECE map of where research risk lives
Research risk sorts cleanly into six types, and the value of the map is that it separates what you can control from what you cannot. A list of things that can go wrong is just anxiety; a taxonomy is a management tool, because each risk type has a distinct owner and a distinct defence.
| Risk type | What it is | Who owns it | Main defence |
|---|---|---|---|
| Independence / conflict | A view bent by banking or trading interests | Research management + compliance | The information wall, disclosure |
| Conduct / distribution | Unfair, selective or non-simultaneous distribution | Compliance + technology | Entitlements, fair-distribution controls |
| Inside information (MNPI) | Publishing or trading on material non-public information | Compliance + surveillance | Restricted lists, wall-crossing, surveillance |
| Accuracy / liability | A wrong or unsupportable call; a factual error | Lead analyst + supervisory | The review gate, triangulation, risk disclosure |
| Commercial / margin | Coverage that cannot justify its cost | Research management / COO | Consumption analytics, productisation |
| Operational / data | A system, data or entitlement failure | IT / data / change | Data governance, resilience, reconciliation |
The first three risks are conduct risks, the last three are business-and-execution risks, and they concentrate at different points in the pipeline. Independence, distribution and inside-information risk all peak at the review-and-publish gate, which is why so much control effort clusters there. Accuracy risk peaks in the analysis and drafting stages, where the view is built. Commercial risk is chronic rather than acute — it is the slow pressure of a coverage list that must earn its cost. And operational risk lurks in the platform, ready to surface at the worst moment. Naming them separately is what lets a research business defend each one with the right control rather than hoping a single policy covers everything.
How to tell a good research business from a lucky one
A single correct call proves nothing; a good research business is visible only across many calls and many clients. The temptation is to judge research by whether its last big call was right, but that is the research equivalent of judging an IPO by its first-day pop — a seductive, misleading single number. Five measures together give the real picture.
| KPI | What it measures | Why it matters |
|---|---|---|
| Hit rate | Share of calls that were directionally right over time | The analyst's actual predictive skill, across many calls |
| Consumption | Readership, interactions, meetings used, per client | Whether clients actually value the coverage |
| Client votes | Rankings in buy-side surveys (e.g. Institutional Investor) | Independent, client-weighted reputation |
| Flow attribution | Trading flow linked to research consumption | Whether research drives the revenue it is meant to |
| Revenue per analyst | Attributed value against fully-loaded cost | Whether the coverage earns its keep post-unbundling |
The trap is to over-weight the hit rate and forget it must be judged against risk and consistency. An analyst who makes one spectacular, lucky call and a dozen poor ones has a great story and a bad record. A hit rate is only meaningful across a large number of calls, adjusted for how bold each was — a Buy that was right in a rising market proves less than a contrarian Sell that was right against the trend. And a high hit rate that no client reads or trades on generates no revenue. The mature view triangulates all five KPIs, exactly as an analyst triangulates a valuation: no single measure is trusted alone, and the business is judged where they agree.
Three real failures define the three ways research goes wrong
The modern research rulebook was not written in the abstract; it was written in response to three real failures, each a different failure mode. One was a failure of independence, one of fair distribution, and one of economics. Together they are not a random selection of scandals — they are the three ways the business breaks.
| Case | Failure type | Where it maps | The lesson |
|---|---|---|---|
| 2003 Global Analyst Research Settlement | Independence / conflict | The wall (Steps 1 and 4) | Research shaped by banking interests is worthless and illegal |
| Goldman Sachs "huddles," 2012 | Conduct / distribution | Fair distribution (Step 5) | Selective disclosure of research is a breach even without a wrong call |
| MiFID II unbundling, 2018 onward | Commercial / structural | The economics (the whole model) | When research must be priced, most of it turns out to be unvalued |
The wall that failed: the 2003 analyst scandal
The founding scandal of the modern business was a systematic failure of independence, and it produced the wall as we know it. During the dot-com boom, star analysts at major banks published glowing ratings on companies while privately disparaging them, because those companies were, or might become, lucrative investment-banking clients. Research had become marketing for the banking division. In April 2003, ten of the largest US investment firms settled with the SEC and the exchanges for 1.4 billion dollars — 387.5 million in restitution to harmed investors and 487.5 million in penalties. Two of the era's most famous analysts, Merrill Lynch's Henry Blodget and Salomon Smith Barney's Jack Grubman, were fined and banned from the securities industry for life. The settlement forced banks to sever the link between research and banking: analysts could no longer be paid for banking work or take part in deal pitches.
Every element of the independence machinery in this module traces back to that settlement. The disclosure of rating distributions, the ban on banking influence over ratings, the compensation firewall, the quiet periods — these are not abstract good practice; they are the specific remedies imposed after research was caught working for the wrong master. When Part 2 called the compliance gate "scar tissue," this is the wound.
The distribution that leaked: the 2012 huddles
The second failure shows that research can breach the rules without a single wrong call — by reaching the wrong clients first. Between 2006 and 2011, Goldman Sachs ran weekly "huddles" in which analysts shared their best short-term trading ideas, and through a programme known as the Asymmetric Service Initiative it passed those ideas to a select group of clients and to its own traders ahead of publishing the research more broadly. The ratings may have been perfectly good; the problem was that favoured clients got them first. In 2012 the firm paid 22 million dollars to settle SEC and FINRA charges that it lacked adequate controls to prevent the misuse of material, non-public information about upcoming research changes.
This is the fair-distribution rule of Part 2, learned the hard way. The huddles case is why the simultaneous-release mechanism is engineered so carefully and why the fair-distribution rule is treated as an absolute rather than a courtesy. A selective tip is a breach even when the underlying view is sound, because the harm is not a bad call — it is an uneven playing field. The lesson is that distribution is a conduct risk in its own right, independent of whether the research was any good.
The economics that broke: unbundling
The third failure was not a scandal but a structural shock, and it revealed that most research was economically unsupported. When MiFID II forced research to be paid for separately from January 2018, the buy-side had to put an explicit price on something that had always been free — and discovered how little of it they valued. The largest European equity investors cut external research budgets by around nineteen percent in the first year. Analyst headcount fell, small- and mid-cap coverage collapsed, and mainstream research commoditised toward zero. This was not a failure of any one bank; it was the failure of the whole free-research model once its cost was made visible. Europe and the UK have since partly reversed course — the EU Listing Act and the UK's joint-payment option — but the strategic lesson stands.
Five technologies genuinely change the research business
Not every "innovation" in research matters; five do, and each attaches to a specific part of the pipeline. The test is simple and worth applying ruthlessly: does the technology change the economics or the risk of the business, or does it just change how research looks? Five pass the test.
| Trend | Where it attaches | Why it matters |
|---|---|---|
| AI-assisted production | Analysis and drafting (Steps 2–3) | Compresses the mechanical work, freeing time for the differentiated view |
| Digital distribution & consumption analytics | Distribution and value capture (Steps 5–7) | Meets clients everywhere and makes value provable |
| RegTech automation | The compliance gate (Step 4) | Automates disclosures, entitlements and record-keeping |
| Cloud and API-first architecture | The whole platform | Joins the fragmented chain and delivers structured output |
| Structured, machine-readable output | Distribution (Step 5) | Reaches quant clients who cannot consume prose |
The first three attach to specific stages and change the cost or risk there. AI-assisted production compresses the mechanical work of digesting results, updating models and drafting, which lowers the cost of production and frees the analyst for the differentiated view — a direct economic change. Digital distribution with consumption analytics meets clients across every channel and, crucially, makes research value provable, which is the difference between defending a budget and losing it. RegTech automation attacks the manual compliance overhead at the gate — assembling disclosures, enforcing entitlements, keeping records — turning a heavy manual burden into a controlled system.
The last two are architectural, and they enable the rest. Cloud and API-first architecture is what joins the fragmented six-block chain end to end, so authoring, distribution and CRM finally see one another; without it, the other trends stall on integration. And structured, machine-readable output is what lets research reach the quant and systematic buy-side at all — a category that is invisible to a bank still shipping only PDFs. Each of the five moves cost or risk. A sixth candidate, the endless stream of prettier dashboards and formats, mostly changes appearance, and the discipline of the test is to spend on the five and not the sixth.
Apply the filter honestly and most "research innovation" fails it. A new note template that looks sharper does not change the economics or the risk — it changes the appearance. A consumption-analytics platform that lets the business price research from evidence changes the economics. An entitlements engine that prevents a distribution breach changes the risk. The five trends above survive the filter because each one demonstrably moves cost or risk; the test is what stops a transformation budget from being spent on decoration. Before funding anything, ask which number on the profit-and-loss or which line on the risk register it moves — and if the answer is neither, it is decoration.
Generative AI earns its place only where it changes cost or risk
Generative AI is the fastest-moving technology in research, and the honest assessment separates real gains from hype. Large language models are genuinely useful at a specific set of tasks, all of which compress the mechanical work around the view without touching the view itself.
| Use of GenAI | The gain | The governing risk |
|---|---|---|
| Summarisation | Digest filings and earnings calls in seconds | Accuracy — a wrong summary corrupts the analysis |
| First-draft generation | Produce a note draft in house style to refine | Hallucination — the draft may state things that are false |
| Data extraction | Pull structured numbers from filings into models | Extraction errors cascade through the model |
| Q&A over the research corpus | Answer questions across the bank's whole library | Confidentiality — the corpus contains MNPI |
| Personalisation & search | Surface the right research to the right client | Fair distribution must still hold |
Where AI helps is real and already in use, but it stops precisely at the analyst's judgement. During peak earnings season, an analyst covering twenty stocks can use a model to summarise each result and transcript in minutes, flag surprises against her forecast, pull the reported figures into her model, and draft the note in house style — then she rewrites the thesis, owns the call, and sends it through the same compliance gate as always. The mechanical work compresses; the differentiated view, which is the entire reason research has value, stays human. That is not a limitation to be engineered away — it is the point.
The risks are specific, and they are why European banks deploy AI under tight governance. Published research must be right, so hallucination and inaccuracy are not tolerable and every AI output must be checked, with the analyst remaining accountable. The research corpus contains material non-public information, so a Q&A tool over it is a confidentiality and MNPI risk that must be walled and access-controlled. Model-risk governance requires the AI's behaviour to be explainable and logged. And there is a commercial risk: if everyone can auto-generate a summary, summaries are worth nothing, and only genuinely differentiated insight retains value. Under GDPR and the emerging EU AI Act, European deployment is controlled and human-in-the-loop by design.
The pain points practitioners actually feel
Behind the risk taxonomy sit the day-to-day frustrations that make research a rich, ongoing change domain, and they cluster around three themes. Ask anyone running a research business what hurts, and the answers are remarkably consistent: proving value, feeding the machine with good data, and joining up a fragmented stack — all under economic pressure. These are not abstract risks but the concrete problems a transformation is hired to solve.
| Pain point | What it feels like | Root theme |
|---|---|---|
| Proving & pricing value | Attributing worth to intangible ideas and access | Value measurement |
| Margin & budget pressure | Shrunken budgets, commoditised content, hard coverage cuts | Economics |
| Fragmented systems | Authoring, distribution, CRM and analytics that don't join up | Integration |
| Compliance & entitlements | Heavy, error-prone, largely manual disclosure and distribution | Controls |
| Data sourcing & quality | Costly, complex sourcing, cleaning and licensing of data | Data |
| Staying relevant & AI-ready | Competing with in-house, IRPs, alt-data; adopting AI safely | Strategy |
The pattern is that most pain points are about value measurement, data, and integration under economic pressure. Proving research's worth requires analytics the business often lacks. Feeding it good data requires a foundation that is expensive to build. Joining up a fragmented stack requires the integration that fell by the wayside when systems were bought piecemeal over years. Solving these is precisely where research transformation creates value — and where AI is now a major lever, because it attacks the cost of production and the drudgery of data-gathering at the same time.
Two applied cases show the transformation agenda in practice
The live policy debate: rebundling, and why it is not a full reversal
Europe and the UK are in the middle of partly undoing unbundling, but the reversal is deliberately incomplete, and that matters. Having watched SME coverage collapse, regulators moved to bring bundled payment back — but neither the EU Listing Act nor the UK's joint-payment option is a return to the pre-2018 world. Both keep guardrails: research costs must remain identifiable and separable even when paid jointly, budgets cannot simply scale with trading volume, and clients must be told what they are paying for. The point is to restore coverage without restoring the conflict — to let a bank be paid for research on a small company again without letting "free" research quietly induce over-trading.
| Regime | What changed | The guardrail kept |
|---|---|---|
| EU Listing Act | Re-permits bundled payment; threshold removed from 2026 | Transparency to the end client |
| UK FCA joint payment | Bundled payment option from 2024, extended to funds in 2025 | Separately identifiable research charges |
| US regime | Never unbundled; SEC/FINRA rules govern conflicts | No change; cross-border friction persists |
The result is a fragmented map that cross-border research businesses must navigate at once, and that fragmentation is itself a cost. A bank serving European, UK and US clients now operates under three subtly different rulebooks on how research may be paid for, entitled and distributed. The systems that track entitlements and payments — already the most heavily engineered part of the platform — must flex to accommodate all three, plus the transitional states as the rules change. The strategic imperative that unbundling created has not gone away: differentiate, price, measure, automate. Rebundling softens the economic squeeze on small-cap coverage; it does not return research to the comfortable invisibility it enjoyed before 2018.
Running research as a business: the commercial operating model
Post-unbundling, a research desk is run like any other cost centre that must earn its keep, and that demanded a commercial operating model it never had before. For decades research was managed as a craft — cover what matters, hire good analysts, publish. Today it is managed as a business, with a profit-and-loss discipline, a client-value function, and hard decisions about where to spend a fixed budget. Four commercial disciplines define the modern operating model, and together they are the survival toolkit.
| Discipline | The question it answers | The tool it uses |
|---|---|---|
| Coverage economics | Which coverage actually pays for itself? | Consumption and revenue attribution |
| Client value management | Which clients value us, and how much? | Account tiering, the value dialogue |
| Productisation & tiering | How do we package and price the offering? | Service tiers, premium à la carte |
| Cost & automation | How do we produce more with less? | Data feeds, workflow, AI |
Coverage economics is the hardest and most political discipline, because it means dropping coverage clients used to get for free. When every name in the universe must earn its cost, the business has to confront which sectors and stocks generate enough consumption and flow to justify the analyst time, and which do not. The answer is usually to concentrate: drop low-value coverage nobody pays for, and double down on the few sectors where the bank is a genuine top-three voice. This is painful — it means visible retreat from names the bank once covered — but a research business that spreads a shrunken budget evenly across everything ends up differentiated in nothing.
Client value management is the mirror image: not which coverage pays, but which clients do. The business tiers its accounts by the value they represent, directs scarce analyst time toward the clients who value and pay for it, and runs the annual value dialogue from consumption evidence. The through-line across all four disciplines is the same shift unbundling forced everywhere: from a craft measured by output to a business measured by value. The desk that internalises this survives; the one that keeps managing research as a cost of doing business does not.
Stepping back: the transformation roadmap and where to invest first
Seen from the altitude of a decision-maker rather than an engineer, the technology agenda for research is not a list of systems but a sequence — and the order matters more than the tools. The detailed platform of Part 4 can overwhelm; the strategic view is simpler. There is a natural order in which to invest, because each step is the precondition for the next. Skip the sequence and money is spent on capabilities that cannot yet deliver value.
| Priority | What to fix | Why it comes first |
|---|---|---|
| 1 | Join the fragmented chain | Nothing downstream works while systems don't talk |
| 2 | Measure consumption end to end | You cannot price or defend what you cannot see |
| 3 | Automate compliance and production | Frees cost and reduces the biggest conduct risk |
| 4 | Deliver structured, multi-channel output | Reaches clients — including quants — where they are |
| 5 | Add AI, under governance | Compresses the mechanical work, once the data is ready |
The sequence is not arbitrary; each step unlocks the one after it. Joining the chain has to come first, because consumption analytics, structured delivery and AI all depend on systems that talk to each other. Measuring consumption comes second, because pricing, targeting and budget defence all rest on it. Automation comes third, attacking both cost and the conduct risk at the compliance gate. Structured, multi-channel delivery comes fourth, opening the platform to clients who cannot consume prose. And AI comes last — not because it matters least, but because a language model over a fragmented, poorly-governed data foundation produces confident nonsense. The order is the strategy.
The buy-versus-build decision, seen at portfolio altitude, is a single principle applied consistently: buy the commodity, build the edge. Buy the RMS, the terminals, the foundation models — these are solved problems where no bank can out-build a specialist. Build, or heavily customise, the consumption analytics, the pricing engine, and the AI guardrails — because these encode the bank's own commercial judgement and competitive advantage. Every technology choice in research reduces to asking which side of that line a capability sits on. Get the line wrong — build a commodity, or buy your differentiation — and you have either burned money reinventing a terminal or handed your edge to a vendor who sells it to your competitors too.
The value at stake is not efficiency for its own sake but the survival of the franchise, and that reframes the whole investment case. A research business that joins its chain, measures its value, automates its drudgery and adopts AI safely can prove its worth, price it, and produce it more cheaply — which is precisely what a priced, competitive, post-unbundling market demands. The transformation is not a cost-reduction programme dressed up as strategy; it is the difference between a research business that can demonstrate it earns its cost and one that quietly cannot. That is the altitude at which the IT agenda should be judged: not which system is newest, but whether the platform lets the business prove it deserves to exist.
What makes the six-part risk taxonomy a management tool rather than a list?
Each risk type has a distinct owner and defence, and the map separates the highly-controllable conduct risks (independence, distribution, MNPI) from the partly-controllable accuracy risk and the strategic commercial risk, telling the business where to spend.
Why is judging research on its last big call a fallacy, and what should replace it?
A single call can be luck; predictive skill shows only across many calls, risk-adjusted, and only matters if clients consume and trade on it. The five KPIs — hit rate, consumption, client votes, flow attribution, revenue per analyst — must be triangulated.
Name the three real failure cases and the failure mode each represents.
The 2003 Global Analyst Research Settlement (independence/conflict), the 2012 Goldman Sachs huddles (fair distribution/selective disclosure), and MiFID II unbundling from 2018 (commercial/structural).
Why was the 2012 huddles case a breach even if the ratings were sound?
Because favoured clients received research changes ahead of others; the harm is an uneven playing field from selective disclosure, independent of whether the underlying call was correct.
What is the filter that separates real transformation from decoration?
Whether the technology demonstrably changes the economics or the risk of the business, versus merely changing how research looks; if it moves neither the P&L nor the risk register, it is decoration.
Where does generative AI help in research, and where must it stop?
It compresses mechanical work — summarisation, drafting, data extraction, corpus Q&A — but must stop at the analyst's differentiated judgement, which stays human; every output is checked and passes the same compliance gate, with the human in the loop as a control.
Why do European banks deploy AI in research under especially tight governance?
Published research must be accurate (hallucination is intolerable), the corpus contains MNPI (a confidentiality risk), and under GDPR and the EU AI Act deployment must be explainable, logged and human-in-the-loop by design.
What are the four disciplines of the modern research commercial operating model?
Coverage economics (which coverage pays), client value management (which clients value us, via tiering and the value dialogue), productisation and tiering (how to package and price), and cost and automation (producing more with less).
In what order should a research business invest in technology, and why does order matter?
Join the fragmented chain, then measure consumption end to end, then automate compliance and production, then deliver structured multi-channel output, then add AI under governance — because each step is the precondition for the next, and AI over a fragmented, poorly-governed foundation produces confident nonsense.
5
Why is judging research by its last big call a fallacy?
Right.
No.
It can, with several KPIs.
No.
Why were the 2012 “huddles” a breach even if the ratings were sound?
Not the issue.
No.
Even sound ratings don't excuse it.
Right. Fairness, not accuracy, was breached.
Which case represents the commercial/structural failure mode?
Independence and conflict.
Unrelated to research.
Right. It changed how research is paid for.
Fair distribution.
Which risks are the most controllable in research?
Partly controllable at best.
Not very controllable.
Right. Accuracy is only partly controllable.
Not controllable.
Module Summary — The Six Ideas That Matter Most
If you keep six things from this module, keep these. Each one has run through every part, and together they are the through-line of the whole research business.
Research is the intelligence layer, and the product defines the platform
Research sits upstream of every trade — it is the reason clients engage and the content that makes a bank worth trading with. It produces nothing you can hold, only an argument, but that argument shapes what clients think and therefore what they do. The kind of research a bank produces — equity, credit, macro, the depth of coverage, the reach of its access — defines the platform it must build, just as the product defined the platform in every other module. Get the product right and the systems follow; confuse the platform for the product and you build tooling around research nobody values.
Data cascades, so an error is never contained
A research note is a structured object, and every downstream system trusts the one before it, so a mistake at the source reaches the client fee. A rating tagged to the wrong entitlement list becomes a distribution breach, then a corrupted consumption record, then a mispriced client. The atom of the business — the note record, with its identifier, rating, timestamp and entitlement tag — has to be right at the point of creation, because nothing downstream can fix an error it inherited and trusted. Data quality in research is not hygiene; it is the difference between a working business and a cascade of failures.
Independence is the product, and it is where the value is really made
Strip out the wall, the gate and the fair-distribution rule, and research still produces notes — but they are marketing, and clients stop paying. The entire expensive, slow apparatus of the business exists to make the view trustworthy: the information barrier that separates research from banking and trading, the compliance gate that makes a note irreversible only after review, and the simultaneous release that gives no client an edge. The value a client pays for is not the note; it is a view they can trust and cannot easily make themselves. Independence is not a constraint on the product. It is the product.
Compliance is core architecture, not an afterthought
The entitlements, the disclosure engine, the restricted list, the immutable audit trail — these are not features bolted onto a publishing tool; they are the structure it hangs on. A research platform cannot be assembled from generic content systems because the compliance-and-entitlements spine runs through every block, from authoring to consumption. The 2003 scandal and the 2012 huddles are what happens when that architecture is weak. Fair distribution, conflict disclosure and inside-information control are load-bearing walls, and a business that treats them as overhead is one incident away from the front page.
Unbundling quietly reshaped every behaviour in the business
Just as league tables distort behaviour in capital markets, the economics of unbundling reshaped everything in research — what gets covered, who gets served, how value is measured. The moment research had to be priced, breadth became a liability, differentiation became survival, and measuring consumption became business-critical. Coverage concentrated, mainstream content commoditised, and the whole business reorganised around proving its worth. The partial reversal — the Listing Act, the UK's joint-payment option — softens the edge but does not undo the lesson: research that cannot show it is valued will not be paid for.
Real transformation moves cost or risk, including GenAI under governance
The five technologies that matter — AI-assisted production, digital distribution and analytics, RegTech, cloud architecture, structured output — each demonstrably move cost or risk; everything else is decoration. Generative AI belongs on that list only where it compresses mechanical work and only under tight governance, with the human in the loop as a control rather than a limitation. The differentiated view stays human because it is the one thing clients pay for. The filter is unforgiving and worth applying to every proposal: which number does this move, on the profit-and-loss or the risk register? If neither, it is decoration.
Process Cheat Sheet — The Whole Business on One Page
A single reference for the entire research and advisory flow — from an analyst's idea to a client acting on it, and the numbers and rules that frame the business.
The production pipeline at a glance
| Stage | Timing | Core activity | Key output | Lead & systems |
|---|---|---|---|---|
| 1. Idea & coverage | Ongoing | Spot a mispricing or theme in coverage | A variant view worth publishing | Analyst; authoring platform |
| 2. Analysis & modelling | Days–weeks | Build the model, triangulate a target | Rating and price target | Analyst + RA; modelling tools |
| 3. Draft the note | Days | Write thesis, model, catalysts, risks | The draft note | Analyst + RA; RMS |
| 4. Review gate | Hours–days | Supervisory + compliance sign-off | An approved, disclosed note | Supervisory analyst + compliance; workflow |
| 5. Publish | Minutes | Simultaneous release to entitled clients | The published note | Publishing & entitlements engine |
| 6. Distribute & market | Days–weeks | Market the call; two client routes | Institutional flow; suitable advice | Research sales, advisers; CRM |
| 7. Maintain & follow | Ongoing | Update the view; track the call | Track record; consumption data | Analyst; consumption analytics |
Key numbers and rules of thumb
| Metric | Rule of thumb |
|---|---|
| Analyst coverage universe | ~15–25 stocks or a defined market |
| Price target horizon | Usually 12 months |
| Valuation methods triangulated | At least two (DCF + multiples), often three |
| Rating scale | Three tiers: Buy / Hold / Sell (or Over/Neutral/Underweight) |
| Distribution rule | Fair and simultaneous to all entitled clients — a hard line |
| MiFID II unbundling live | 3 January 2018 |
| EU equity research budget cut, year one | ~19% at the largest European investors (illustrative of the shock) |
| Global Analyst Research Settlement (2003) | $1.4bn, 10 firms, two analysts banned for life |
| Goldman "huddles" settlement (2012) | $22m, selective distribution of research changes |
| EU Listing Act re-bundling | Adopted 2024; threshold removed from June 2026 |
| UK FCA joint-payment option | Live from August 2024 |
| Suitability duty | Attaches only to personal advice, never to general research |
The one-sentence version
Research manufactures independent, defensible views and sells them — fairly, simultaneously, and provably — to clients who could not easily make them, while advisory turns those views into recommendations tailored and documented to fit one client at a time; the whole business balances being commercially viable against being trustworthy, and lives or dies on both.