Skip to content
Corporate & Investment Banking My progress Français

Home / Capital Markets & Securities Services / Capital Markets / Module 5

5Client Onboarding and KYC in Global Markets

From First Contact to Trade-Ready · How a Markets Counterparty Is Assessed, Documented, and Switched On

Who is this client, and are we allowed to trade with them?

Reading time
2 h 30
Sections
57
Recall questions
53
Quiz questions
20

Onboarding is the gate every trade passes through, and it is the least understood part of the bank

Every module in this series describes a business that makes money by executing transactions. None of those transactions can happen until the client on the other side has been onboarded. That makes onboarding the one process that sits upstream of everything else, and it is routinely treated as administrative overhead by the people who depend on it most.

Module 1 on Equity Capital Markets gave this process a single paragraph. In the section on the internal gauntlet a deal must pass, it noted that the client must be onboarded first, that know-your-customer checks establish who the company and its owners really are, and that slow onboarding is a common and unglamorous cause of deal delay. That paragraph is the door this module opens. Everything the product modules describe happens on the far side of it.

A trading relationship does not begin with a trade. Before a single price is quoted, the bank must satisfy itself on five separate questions, and each one is owned by a different function. May the bank lawfully face this entity at all, from this booking entity, in this country, for these products? Who is this client, really, and does dealing with them expose the bank to financial crime? Can the bank legally enforce its contracts against them if things go wrong? Does the bank have appetite for the credit exposure the relationship will create? And does the operational plumbing exist to confirm, margin, and settle what they trade? Onboarding is the process that answers all five. Its output is a status, not a document: the client becomes trade-ready.

The cost of that status is chronically underestimated. A complex markets counterparty routinely takes four to nine months from first request to first trade, consumes effort across six or more internal teams, and produces a client record with several hundred populated fields. None of that generates revenue. All of it determines whether revenue is legally collectable, correctly margined, and accurately reported for the following decade.

Markets onboarding and corporate onboarding solve genuinely different problems

This distinction matters enough to state at the outset, because conflating the two is the most common error made by people arriving from adjacent parts of the bank.

Corporate banking onboarding is about a borrower. When coverage onboards a corporate client, the organising question is creditworthiness over time: can this company service debt, what does its business actually do, who controls it, and what is the relationship worth across the whole bank? The unit of analysis is the company. The documentation is a facility agreement. The relationship manager owns the client, and the process is oriented around a lending decision that will be reviewed annually.

Markets onboarding is about a counterparty and a permission set. When Global Markets onboards a client, the organising question is different: under which legal agreement, for which products, in which jurisdictions, against which collateral terms, may this specific legal entity face this specific bank entity? The unit of analysis is not the client. It is the combination of legal entity, product, and jurisdiction, and a single commercial relationship can generate dozens of those combinations. A corporate onboarding produces a credit file. A markets onboarding produces a matrix of entitlements.

The consequences of the difference run through everything that follows. Markets clients are frequently agents rather than principals, so the entity that signs is not the entity that bears liability. Markets exposure is contingent and moves with the market rather than being drawn down on a schedule, so the credit assessment is a distribution rather than a number. Markets contracts depend for their capital treatment on whether close-out netting is legally enforceable in the counterparty's jurisdiction, which is a question corporate lending rarely has to ask. And markets clients must be classified under trading regulation, which corporate borrowers largely escape. Coverage onboarding asks whether the bank wants the client. Markets onboarding asks that too, then asks four further questions the corporate side never reaches.

Two threads run through everything: the business and the technology

The business thread follows the decisions. Who assesses what, in which order, with what authority to stop the process, and where the commercial terms are actually settled. Onboarding looks procedural from outside and is deeply negotiated from inside; the credit and collateral terms agreed during documentation have a measurable effect on what the relationship costs the bank for its entire life.

The technology thread follows the record. Onboarding is, from a systems perspective, the creation and continuous maintenance of one authoritative object: the client record. Everything downstream reads from it. Margin engines read its collateral terms. Netting calculations read its legal agreements. Regulatory reports are keyed on its identifiers. Sanctions screening runs against its names and ownership. The record is created once, badly or well, and the consequences of that choice are permanent.

How to read this module

Read it as a narrative, not a reference card

The parts build on each other deliberately. Part 1 establishes what the process is for and how the different onboarding routes compare. Part 2 walks the full end-to-end flow and is the longest and most important part of the module. Part 3 maps who does what and where the independence lines run. Part 4 opens the systems and the data. Part 5 covers what goes wrong, what the failures have actually cost, and where the technology is genuinely changing.

A capstone exercise follows Part 5, presenting an unfamiliar client structure and asking the reader to scope the onboarding, identify the gates, and predict where it will stall. A worked solution follows it. The self-check boxes test recall; the exercise tests whether the module has actually transferred judgement. A cheat sheet and a glossary close the module for later reference.

One fictional client runs through the whole module. Northwind Asset Management is introduced in Part 2 and reappears at every stage, so the abstract process always has a concrete case attached to it. Northwind is invented, and every figure attached to it is illustrative only.

Each coloured box plays a fixed role

Seven box types recur, and each one signals a specific kind of content. Key insight crystallises an idea that the surrounding prose has been building toward. Going deeper adds technical depth that a first reading can skip. IT and data lens identifies the systems and data structures that carry the process being described. Watch out flags a common misunderstanding or a live risk. Worked example grounds the mechanics in numbers, usually through Northwind. In a nutshell recaps a part in a few sentences. Check your understanding closes each part with questions and answers, and is best used by covering the answers first.

What you will master

You will understand the onboarding process end to end

You will be able to walk the full sequence from client request through cross-border permissibility, financial-crime due diligence, credit assessment, legal documentation, static data setup, connectivity, and activation, and explain what each stage produces, who owns it, how long it takes, and why it cannot be skipped or reordered.

You will be able to map every actor

You will know which internal teams touch an onboarding file, which of them can unilaterally stop it, where the independence line between revenue generation and financial-crime control runs and why it exists, and which external parties, from clearing houses to sanctions data vendors, participate in the process.

You will understand where the credit and collateral terms actually come from

You will be able to explain how a threshold, a minimum transfer amount, a haircut schedule, and an initial margin requirement are arrived at, why they are argued rather than quoted, and how a netting opinion in a distant jurisdiction changes the capital the bank must hold against the relationship. You will also be able to trace those terms forward into the price the client is quoted, through the valuation adjustments they create.

You will be able to trace client data front to back

You will be able to follow a single client record from intake through screening, credit, documentation, and reference data into regulatory reporting, name the systems it passes through, and explain precisely how one incorrect classification field becomes a wrong margin call, a broken netting set, and a rejected regulatory report.

You will recognise the risk and transformation landscape

You will know the six risk types the process is built to contain, the metrics that separate a well-run onboarding function from a fast one, three real failures and what each cost, and where perpetual KYC, automation, and generative AI genuinely change the economics rather than merely the appearance.

1Context and the Big Picture

A bank's obligations begin when it accepts a client, not when it trades

Start from the legal position rather than the process, because the process only makes sense once the obligations are clear. Three separate bodies of law converge on the moment a bank agrees to deal with someone, and each one imposes a duty that must be discharged before the first transaction.

Anti-money-laundering law makes the bank responsible for knowing who it deals with. Across every major jurisdiction, a regulated financial institution must identify its customer, verify that identity against reliable independent evidence, understand the ownership and control structure behind it, form a view of the money-laundering and terrorist-financing risk the relationship presents, and monitor it for as long as it lasts. This is customer due diligence, universally abbreviated to CDD. It is not advisory guidance. Failure is a criminal and regulatory matter, and the fines discussed in Part 5 run into billions.

Sanctions law makes the bank responsible for who it deals with, on pain of strict liability. Sanctions regimes prohibit dealing with designated persons, entities, and in some cases entire sectors or territories. Unlike anti-money-laundering rules, which are risk-based and allow judgement, sanctions compliance is largely absolute: a prohibited transaction is a breach whether or not the bank knew. Screening the client, its owners, and its controllers against sanctions lists is therefore not a risk assessment but a hard gate.

Prudential and contract law make the bank responsible for what the relationship costs it. A trading relationship creates credit exposure that moves with the market. The bank must hold regulatory capital against that exposure, and the amount depends on whether its contracts are enforceable, whether close-out netting works in the client's jurisdiction, and what collateral it can demand. These are not compliance questions. They are balance-sheet questions, and they are settled during onboarding.

Five questions, five owners, one sequence

The process exists to answer five questions, and no single function can answer them all. May the bank lawfully face this client at all, from this booking entity, in this country, for these products, owned by legal and cross-border compliance. Who is this client and what financial-crime risk do they carry, owned by the financial-crime function. What is the bank's appetite for the exposure, owned by credit risk. Under what enforceable terms will the parties deal, owned by legal and documentation. And can the operation actually confirm, margin, and settle the resulting trades, owned by onboarding operations. Because the answers depend on each other, the questions must be asked in order. Credit cannot size an exposure before it knows which legal entity it faces. Legal cannot draft a schedule before credit has set a threshold. Operations cannot open an account before the documents exist.

The first question is the one most often assumed rather than asked, and it is the cheapest to get wrong. It is treated in full below, because it determines whether any of the other four are worth answering.

That dependency is why onboarding is slow, and why parallelising it is harder than it looks. Every transformation programme in this space begins by proposing to run the stages concurrently. Some of it works, and Part 5 covers where. But the core sequence is a genuine dependency chain, not an accident of organisation design.

Markets onboarding is the operating spine of the trading division

A bank is many businesses under one roof, and they onboard differently. Retail banking onboards individuals at volume with automated identity checks. Corporate and commercial banking onboards borrowers through coverage teams, oriented around a credit relationship and a facility agreement. Global Markets onboards counterparties, and the difference is structural rather than cosmetic.

The markets client is often not a single entity, and often not the entity that bears the risk. An investment manager signs agreements on behalf of funds it manages. A pension scheme trades through a delegated manager. A corporate treasury hedges for subsidiaries in several countries. In each case the entity the bank negotiates with is an agent, and the entities the bank is actually exposed to are the underlying principals. Every principal needs its own identity verification, its own credit line, its own set of documents, and its own identifier. One commercial conversation becomes many onboarding files.

Client Lifecycle Management orchestrates the process without owning the decisions

The team in the middle is usually called Client Lifecycle Management, or CLM. It sits in the Global Markets chief operating officer function, not in compliance and not in the front office. Its job is to run the process: collect documents from the client, route the file to each reviewing function, chase the blockages, maintain the record, and tell sales when the client is live. It is a coordination function with a data mandate.

CLM decides nothing that matters, and that is deliberate. It cannot clear a sanctions hit, set a credit limit, or agree a contractual term. Those decisions belong to functions with independent authority, for reasons Part 3 develops. What CLM owns is the completeness and accuracy of the client record, which turns out to be the most durable asset the process produces.

Whether the bank may face the client is a different question from whether it wants to

Commercial teams reason about clients. Regulators reason about legal entities operating in territories. Before any due diligence begins, someone must establish that the specific bank entity proposed can lawfully provide the specific products to a client located in that specific country. The answer is frequently no, and discovering it late is expensive in a way that is entirely avoidable.

A branch and a subsidiary are not interchangeable

A branch is part of the same legal entity as its parent. It has no separate capital, its obligations are obligations of the whole bank, and it generally operates under the parent's home authorisation extended by host-state permissions. A client facing a branch is facing the parent bank.

A subsidiary is a distinct legal person, locally authorised, with its own capital and its own balance sheet. A client facing a subsidiary has no claim against the parent unless a guarantee says otherwise.

The choice changes almost everything downstream. It determines which netting opinion applies, which regulator supervises the relationship, which capital regime the exposure consumes, which entity appears in transaction reports, and whether exposures to the same client through different arms of the bank can be aggregated or must be managed separately. This is why the scoping question in stage one asks which of the bank's own entities will book the trades, and why the answer is a legal determination rather than a preference.

Licensing determines what may be offered, to whom, and where

Authorisation is granular. A licence permits defined activities, for defined client categories, in a defined territory. A bank entity licensed to deal in one jurisdiction may have no permission to market derivatives in another, or may market to institutional clients while being barred from anything resembling a retail approach.

Within the European framework, firms authorised in one member state have historically been able to serve clients across the bloc under passporting arrangements. Outside that framework, access depends on third-country regimes that vary considerably by jurisdiction and by client category, sometimes requiring an equivalence determination and registration, sometimes requiring local presence, and sometimes prohibiting cross-border provision entirely. The departure of the United Kingdom from the European Union converted a large volume of previously routine cross-border business into exactly this question, which is why booking-model work has occupied so much of the industry's attention.

Reverse solicitation is narrower than commercial teams believe

Where a client approaches the bank entirely on its own initiative, some regimes permit the bank to respond without local authorisation. This is reverse solicitation, and it is genuine but tightly bounded. Supervisors have repeatedly warned that it cannot be manufactured, that it does not survive marketing or relationship outreach that preceded the approach, and that it does not extend to product categories beyond what the client specifically requested.

Relying on it as a business model rather than as an occasional exception is a recognised supervisory concern. The control is that the determination is made and documented by legal or compliance, not asserted by the desk that wants the trade.

Seven onboarding routes cover the entire field

Not every client needs everything. A pension fund that wants to buy shares needs far less than a hedge fund that wants to run leveraged cross-product positions. The bank therefore operates several distinct onboarding routes, defined by the product the client wants and the master agreement that governs it. Ordering them by elapsed time reveals the underlying logic immediately.

Route Core documentation Typical elapsed time Complexity Defining question
Cash equities, FX spot Terms of business, MiFID categorisation 2-4 weeks Low Can we settle against this account?
Listed and cleared derivatives Give-up agreement, clearing mandate 4-8 weeks Low-medium Who clears, and who carries the default risk?
Repo GMRA plus annexes 6-12 weeks Medium Whose collateral, valued how?
Securities lending GMSLA plus annexes 6-12 weeks Medium What can be lent, and against what margin?
Cleared OTC derivatives ISDA plus clearing addendum 8-16 weeks Medium-high Does the CCP accept this client?
Bilateral OTC derivatives ISDA Master, Schedule, CSA 12-24 weeks High Is netting enforceable, and at what threshold?
Prime brokerage PB agreement, margin lending, custody 16-30 weeks Very high What happens if the client fails at 3am?

Elapsed times are indicative ranges observed across the industry and vary widely by institution, jurisdiction, and client readiness; treat them as orders of magnitude rather than benchmarks.

Execution-only cash products ask the smallest question

Buying a listed share on an exchange creates almost no lasting exposure. The trade settles in two business days against delivery of cash, and once it has settled the bank and the client have no continuing obligation to each other. There is credit risk, but it lives for forty-eight hours and is covered by delivery-versus-payment settlement, described in Part 2. What the bank needs is identity, a regulatory categorisation, an account, and settlement instructions. Nothing needs to be negotiated, which is why these routes complete in weeks.

FX spot occupies an unusual position. It settles quickly and, in the European framework, spot foreign exchange is not treated as a financial instrument at all, which removes a layer of trading regulation. FX forwards and swaps are a different matter: they create exposure over time, so they fall under derivatives documentation and follow the ISDA route.

Listed derivatives move the credit question to a clearing house

A futures contract lives for months, so someone must stand behind it. The resolution is central clearing. When a futures trade is executed, it is novated to a central counterparty, or CCP, which becomes the buyer to every seller and the seller to every buyer. The client's credit exposure is then to the CCP rather than to the executing bank, and the CCP protects itself by demanding margin daily.

The bank's onboarding question changes shape accordingly. It is no longer whether the bank wants long-term exposure to the client, because it largely does not have any. It is whether the client has a clearing arrangement, who the clearing broker is, and how executed trades reach that clearing broker. The instrument that answers the last question is the give-up agreement, a tri-party document among client, executing broker, and clearing broker that establishes the executing broker's right to hand the trade over for clearing. The industry standardised this on the FIA give-up template and administers it electronically.

Repo and securities lending are collateral businesses, so the documents are about collateral

A repurchase agreement is a secured loan dressed as two trades. One party sells securities for cash and simultaneously agrees to buy them back at a set price on a set date. Economically it is borrowing against collateral. Legally it is a sale and a repurchase, and that legal form is the point, because ownership of the securities genuinely transfers, which is what makes the lender safe if the borrower fails.

Securities lending inverts the flow. The client borrows securities, typically to settle a short sale, and posts cash or other securities as collateral. Both businesses are governed by industry master agreements, the Global Master Repurchase Agreement for repo and the Global Master Securities Lending Agreement for lending, and in both the negotiation concentrates on what collateral is eligible, how it is valued, what haircut applies, and how quickly margin moves.

Bilateral OTC derivatives are where every hard question appears at once

An uncleared swap can live for thirty years, and its value moves the whole time. Two parties agree to exchange cash flows on terms negotiated between them. Nothing is novated to a clearing house, no exchange stands in the middle, and the exposure each party has to the other changes continuously with interest rates, credit spreads, or currencies. This is the route that requires everything: a master agreement, a negotiated schedule, a collateral annex, legal opinions on enforceability, a regulatory classification for both parties, and initial margin calculated under a prescribed model.

This route is the module's reference case. It is chosen for the same reason the IPO was chosen in the equity capital markets module: it is the most complex instance of the process, and everything simpler is a subset of it. A team that can onboard a client for bilateral derivatives can onboard for anything else by omission.

Prime brokerage is the most operationally demanding route of all

Prime brokerage bundles financing, custody, clearing, and reporting into one relationship. The client, usually a hedge fund, executes with many brokers but settles and finances everything through one prime broker, who lends against the portfolio, holds the assets, and provides consolidated reporting. The onboarding is correspondingly heavy, because the bank is extending secured credit, taking custody of assets, and accepting concentrated risk to a leveraged counterparty.

The defining question is what happens on a bad day. Prime brokerage exposure can move violently, and the bank's protection is its ability to value the portfolio, call margin, and liquidate collateral quickly. Onboarding is where the terms governing all three are set. Part 5 examines a case where those terms proved inadequate.

Onboarding is a cost centre whose real product is time-to-revenue

The direct cost is significant and almost entirely people. Analysts gathering and verifying documents, screening specialists dispositioning alerts, credit officers modelling exposure, negotiators arguing over clauses, operations staff opening accounts. Technology has reduced the cost per file but has not changed its nature, because the residual work is judgement rather than transcription.

The indirect cost is larger and less visible. Every week a client spends in onboarding is a week the bank earns nothing from a relationship it has already decided it wants. When the process takes months, clients trade elsewhere in the meantime and sometimes never arrive. Competitors with faster onboarding win business on that basis alone, which is why cycle time has become a commercial metric rather than an operational one.

Metric Illustrative benchmark
Elapsed time, simple execution-only client 2-4 weeks
Elapsed time, bilateral derivatives client 3-6 months
Elapsed time, prime brokerage client 4-8 months
Internal teams touching a complex file 6-8
Populated fields in a complete client record Several hundred
Share of elapsed time spent waiting on the client Frequently the largest single component
Periodic review cycle, high-risk client Annual
Periodic review cycle, low-risk client Every three to five years

These figures are illustrative only. They are drawn from commonly cited industry ranges to give a sense of scale, not from any specific institution.

Client profitability quietly decides who gets onboarded first

Onboarding capacity is finite, so it is rationed. Every bank runs a queue, and the queue is prioritised. The formal criteria are usually expected revenue, strategic importance, and the cost the relationship will impose on the balance sheet. This is rational. It is also the mechanism through which commercial pressure enters a process that is supposed to be objective.

The distortion is subtle and worth naming. Prioritisation legitimately determines the order in which files are worked. It must never determine the standard applied to them. The distinction is easy to state and genuinely difficult to hold, because the same senior stakeholders who set priorities also have views about outcomes. Part 3 explains the structural defence: the functions that can stop a client do not report to the people who want the client.

Four structural alternatives change the shape of the problem

The standard bilateral route is not the only way to establish a trading relationship, and each alternative removes a different part of the burden.

Central clearing replaces the counterparty question with an infrastructure question

When a trade is cleared, the CCP becomes the counterparty to both sides. The bilateral credit assessment largely disappears, replaced by the CCP's own membership and margin requirements. After the 2008 crisis, regulators made clearing mandatory for the most standardised derivatives precisely to achieve this. For onboarding, clearing substitutes a long negotiation over collateral terms with a shorter process of establishing clearing access, though the client must still be accepted by a clearing member and the CCP.

Executing broker and give-up arrangements separate execution from risk

A client can trade with one bank and clear through another. The executing broker takes the order and works it; the clearing broker carries the position and the margin obligation. This lets the client access execution capability without establishing full credit relationships everywhere, and lets banks offer execution to clients they would not extend clearing credit to. The give-up agreement is what makes the arrangement legally coherent.

Utilities and mutualised due diligence attack the duplication

The same client is onboarded independently by every bank it deals with, and each collects nearly identical documents. The waste is obvious, and the industry has repeatedly attempted to mutualise it through shared utilities where a client uploads its documentation once and permits multiple banks to access it. Some of these have worked well in specific niches, particularly for correspondent banking and for the exchange of tax and derivatives documentation. Broader attempts have had a mixed record, and several ambitious platforms have been wound down.

The reason the idea keeps returning and keeps disappointing is worth understanding. Collecting documents is genuinely mutualisable. Assessing them is not, because the regulatory obligation to form a view of the client's risk sits with each bank individually and cannot be outsourced. A utility can remove the collection burden, which is real but is not where the judgement lives.

Reliance and delegation shift the work without shifting the accountability

Some frameworks permit a bank to rely on due diligence performed by another regulated institution, subject to conditions and to the reliant bank remaining accountable. In practice reliance is used selectively, often within a group or with a well-known regulated counterpart. It reduces duplicated effort. It does not reduce responsibility, and supervisors have been clear that a bank cannot outsource its way out of a failure.

The business is cyclical, but the cycle is regulatory rather than economic

Figure 1 Successive regulatory waves have added permanent data requirements to every client record

Capital markets businesses rise and fall with markets. Onboarding rises and falls with rulemaking. Volumes of new clients follow market activity to some degree, but the shape of the function, its headcount, and its technology budget are driven almost entirely by regulatory change, and the pattern has been consistent since 2008.

Each wave added fields that were never removed

The post-crisis derivatives reforms came first. Dodd-Frank in the United States and EMIR in Europe required counterparties to be classified, trades to be reported, standardised derivatives to be cleared, and dealers to follow prescribed business-conduct rules. Classification alone was a substantial exercise: every client had to be categorised, and the category determined which obligations applied.

Margin rules followed and landed hardest. The uncleared margin rules phased in over several years, extending initial and variation margin requirements to progressively smaller counterparties. Each phase brought a new population of clients into scope, and each in-scope client required repapering, new collateral documentation, custodial arrangements for segregated margin, and a calculation methodology. The final phase in 2022 captured a large number of mid-sized asset managers and funds.

Conduct and transparency rules arrived in parallel. MiFID II required clients to be categorised as retail, professional, or eligible counterparty, with the category driving protections, disclosures, and reporting. It also required detailed transaction reporting keyed on legal entity identifiers, which made identifier quality a first-order concern rather than a housekeeping matter.

Financial-crime rules tightened continuously. Successive European anti-money-laundering directives extended beneficial-ownership transparency, expanded the definition of politically exposed persons, and pushed toward centralised registers. A new European anti-money-laundering authority is being stood up to supervise directly, which will change how consistently rules are applied across member states.

Sanctions became an operational discipline rather than a periodic check. The scale and pace of designations from 2022 onward forced institutions to move from screening at onboarding and at review, to screening continuously against lists that change without notice. That shift, more than any single rule, is what made perpetual monitoring an architectural requirement rather than an aspiration.

The cycle sets the rhythm of technology investment

Onboarding technology is bought in response to regulation, not in anticipation of efficiency. Each wave produces the same sequence: a compliance deadline, a manual remediation programme to meet it, a backlog and a cost spike, and then an investment case to automate what was just done by hand. It is an expensive rhythm, and it explains why so many onboarding estates are a sediment of platforms bought during successive waves rather than a designed architecture.

The current wave is different in kind. Perpetual KYC replaces periodic review with continuous, event-driven refresh. It is the first change in this space justified primarily by cost and risk reduction rather than by a deadline, which is why Part 5 treats it as the genuine transformation among several candidates.

  1. Why does markets onboarding treat the legal entity, product, and jurisdiction combination as the unit of work rather than the client?

  2. A client wants to trade listed futures only. Why is this materially faster than onboarding for uncleared swaps?

  3. What single factor best explains the ordering of the seven onboarding routes by complexity?

  4. Why has mutualising KYC through industry utilities repeatedly under-delivered?

  5. What is the difference between prioritising the onboarding queue and applying a different standard to it, and why does it matter?

  6. Why did the uncleared margin rules create more onboarding work than the clearing mandate did?

  7. Why is a rushed onboarding file described as deferring time rather than saving it?

  8. Why does the choice between facing a client from a branch or from a subsidiary change more than the name on the contract?

  9. A desk argues that no local licence is needed because the client called them first. What is the correct response?

1

  1. What is the unit of work in markets onboarding?

  2. Why is onboarding for listed futures faster than for uncleared swaps?

  3. What best explains how complex an onboarding route is?

  4. Why has mutualising KYC through industry utilities under-delivered?

2The End-to-End Flow

Hold the whole arc in your head before the detail

Eight stages carry a client from a sales request to a first trade. The first four assess and negotiate. The next three build and switch on. The last one never finishes. Each stage produces something specific that the next stage consumes, which is why the sequence resists compression.

Stage Typical duration Output Who leads
1. Intake and scoping 1-2 weeks Defined entity, product and jurisdiction scope Sales with CLM
2. KYC and financial crime 3-8 weeks Verified identity, ownership, risk rating Financial crime (2LOD)
3. Credit assessment 2-4 weeks Approved limits and collateral parameters Credit risk (2LOD)
4. Legal documentation 4-16 weeks Executed master agreements and annexes Legal and documentation
5. Static and reference data 1-2 weeks Client record, accounts, settlement instructions Onboarding operations
6. Connectivity and testing 2-4 weeks Working order, margin and settlement paths Onboarding operations with IT
7. Activation and first trade Days Trade-ready status CLM
8. Ongoing lifecycle Perpetual Current, accurate, monitored client record CLM with financial crime

The effort has two peaks, not one

Figure 2 Onboarding effort concentrates twice: once on the file, once on the contract

Most processes have a single moment of maximum pressure. Onboarding has two, and they are different in character. The first peak sits in financial-crime due diligence, where the work is investigative: chasing ownership chains, resolving screening alerts, and deciding what the evidence supports. The second sits in documentation, where the work is adversarial: two sets of lawyers negotiating terms that will govern the relationship for years.

The trough between them is where files quietly die. Credit assessment is comparatively quick, but it depends on information from stage two and produces parameters that stage four will negotiate. Files stall here more often than anywhere else, usually because a question raised in due diligence was never fully closed and credit will not approve against an open item.

Everything after documentation is construction rather than judgement. Stages five through seven are demanding operationally and are where most avoidable errors are introduced, but they involve almost no discretion. The decisions have already been made. What remains is writing them into systems correctly.

Meet Northwind Asset Management

Figure 3 One commercial relationship generating four legal entities, four KYC files, and four credit lines

One relationship, four entities, four files

The agent-principal structure is the single most important fact about onboarding an asset manager, and it is the reason this route was chosen as the reference case over a simpler corporate or bank counterparty.

Northwind signs as agent. The funds are the principals. When Northwind executes an ISDA Master Agreement, it does so on behalf of each fund separately, and the resulting agreement is between the bank and that fund, not between the bank and Northwind. If the Luxembourg fund defaults, the bank has a claim against the assets of that fund alone. It has no claim against Northwind, and none against the Irish fund.

That legal reality dictates the operational shape of the entire onboarding. Each fund requires its own identity verification and ownership analysis, because each is a separate legal person. Each requires its own legal entity identifier. Each requires its own credit line, sized against its own assets. Each requires its own set of documents, or at minimum its own annex to a shared master. Each gets its own regulatory classification, and as the diagram shows, they do not all get the same one. And each forms its own netting set, meaning exposures can be offset within a fund and never across funds.

The commercial conversation is with one institution. The onboarding workload is four times over, plus the manager itself. Northwind is also assessed, because the bank needs to know who is instructing it even though Northwind carries no trading liability. This asymmetry, one relationship and many files, is what makes asset manager onboarding disproportionately heavy relative to the revenue it produces.

Step 1 — Intake defines the scope, and the scope determines everything downstream

A request arrives from a salesperson, and it is almost always incomplete. The typical form is a message saying a client wants to trade. What it rarely contains is which legal entities will actually face the bank, which products they need, which of the bank's own booking entities they will trade with, and by when. Those four facts define the entire workload, and establishing them is the real content of stage one.

Four scoping questions determine the size of the job

Which legal entities will trade? Not the brand, not the group, but the specific legal persons that will be party to transactions. This is where the Northwind structure is unpacked and four files are created rather than one.

Which products? The product list drives the documentation set. Swaps require an ISDA, repo requires a GMRA, listed futures require give-up arrangements. Adding one product to the scope can add months, and adding a product after documentation has been drafted causes rework that is disproportionate to its apparent size.

Which of the bank's entities? Large banks book trades through several legal entities across jurisdictions, and each combination of client entity and bank entity is a separate contractual relationship with its own enforceability analysis. A European fund facing the bank's London entity and its Frankfurt entity has two relationships, not one.

By when, and driven by what? A client hedging a specific transaction has a hard deadline. A client establishing a general relationship does not. The answer changes prioritisation and tells the bank whether the commercial pressure it is about to experience is real or manufactured.

Permissibility is settled before anything is spent

Once the scope is known, the next question is whether it is lawful. Legal and cross-border compliance confirm that each proposed bank booking entity may provide each proposed product to each client entity in its jurisdiction of establishment. Part 1 sets out why this turns on branch and subsidiary structure, licence scope, and third-country access, and why reverse solicitation is a narrow exception rather than a route around the question.

The output is a permissibility matrix, not a yes or no. For Northwind the answer differs by cell: the London entity may face the Irish and Luxembourg funds for the full product set, while the US pension mandate introduces United States requirements that constrain which bank entity may face it and under which conduct rules. Some cells close, and closing them at week two rather than month four is the entire point of asking early.

New business intake is the first gate, and it is a real one

Before any work is commissioned, the request passes a new business review. This is a light but genuine control that asks whether the bank wants this client at all, before spending money finding out whether it can have them. It screens for obvious prohibitions, jurisdictions the bank has exited, client types outside stated risk appetite, and reputational concerns that would make the answer no regardless of what due diligence finds.

Catching a no at this stage is worth a great deal. The alternative is discovering it in month four, after a screening alert surfaces something that a five-minute check would have revealed at the outset.

Intake creates the first version of the client record, and the quality of that creation echoes for years. The critical operation is entity resolution: determining whether these legal entities already exist somewhere in the bank's systems, possibly under a different name, a legacy identifier, or a former corporate title. Banks that skip this step create duplicate client records, and duplicates are extraordinarily expensive to unwind because downstream systems bind to whichever identifier they were given first.

The intake system is usually the CRM or a dedicated onboarding workflow platform. It should reach out to the legal entity reference database, query external registries by name and identifier, and present candidate matches to the analyst rather than allowing free-text creation of a new client.

Step 2 — Due diligence establishes who the client actually is

This is the stage the module's title refers to, and it is more investigative than its reputation suggests. Know your customer is a poor name for it, because the obligation is not merely to know the customer's name. It is to understand the customer well enough to judge whether transactions they conduct are consistent with what they claim to be.

Identification and verification are different operations

Identification is collecting the claim. Verification is testing it. A client identifies itself by providing its name, registered address, incorporation details, and regulatory status. The bank verifies those claims against reliable, independent sources: company registries, regulatory registers, audited financial statements, certificates of incorporation. The distinction sounds pedantic and is the substance of the control. A file containing everything the client said, and nothing independently confirming it, has performed identification and no verification.

For a regulated entity, verification is comparatively easy. Northwind is authorised by the UK regulator, which maintains a public register confirming its permissions and status. That single source verifies existence, regulatory standing, and the scope of activity it may conduct.

For a fund, verification is harder, because funds are structures rather than operating businesses. The bank must establish the fund's legal form, its domicile, its regulatory status if any, and critically the identity of the parties around it: the management company, the depositary, the administrator, and the auditor. A fund with a reputable depositary and administrator in a well-regulated domicile is a different proposition from one with none of those.

Ownership means following the chain to a human being

The regulatory objective is to identify the ultimate beneficial owner, the natural person who ultimately owns or controls the client. Legal persons cannot launder money on their own behalf. Someone benefits, and the point of ownership analysis is to find out who.

The twenty-five percent threshold, and what it misses

Most frameworks define beneficial ownership by reference to a shareholding threshold, commonly twenty-five percent. Any natural person holding more than that proportion, directly or indirectly, is a beneficial owner and must be identified and verified. Indirect holdings are calculated by multiplying through the chain, so a person owning sixty percent of a company that owns fifty percent of the client holds thirty percent indirectly and is captured.

The threshold is a floor, not a ceiling, and treating it as a ceiling is a classic weakness. Ownership can be deliberately structured to sit just below it. Four unrelated-looking holders at twenty-four percent each may be acting together. The obligation is to understand ownership and control, and a file that stops at the arithmetic has answered a narrower question than the one the regulation asks.

When there is genuinely no beneficial owner

Widely held funds frequently have no owner above the threshold, which is a legitimate outcome rather than a gap. A fund with thousands of investors, none holding more than a fraction of a percent, has no beneficial owner in the shareholding sense.

The analysis then shifts entirely to control. Where no natural person meets the ownership test, frameworks require the bank to identify the senior managing officials who direct the entity. For a fund, that means the board of directors, the management company, and the individuals with authority to instruct. The Northwind funds fall squarely into this category, and their files are built around control rather than ownership.

Control identifies who can actually direct the entity

Ownership and control diverge more often than intuition suggests. A person with no shareholding may control an entity through voting agreements, board appointment rights, veto powers, or contractual arrangements. The bank must identify the individuals who can direct the client, and separately, the individuals authorised to give the bank instructions, which is a narrower and more operational question captured in an authorised signatory list.

Screening tests the client against lists and against the world

Sanctions screening is a hard gate

Every entity and every identified individual is screened against sanctions lists, meaning the consolidated designations published by the relevant authorities in the jurisdictions the bank operates in. A confirmed match ends the relationship. There is no risk-based judgement available and no threshold below which a designated person becomes acceptable.

Screening runs on names, and names are a bad key. Transliteration, ordering, abbreviation, and common surnames all generate matches that are not matches. Screening engines therefore use fuzzy matching, which deliberately over-matches to avoid missing a true hit, and produces a large volume of alerts that must be dispositioned by a human. The tuning of that fuzziness is one of the more consequential technical decisions in the estate.

Politically exposed persons are a risk category, not a prohibition

A politically exposed person holds or has held a prominent public function, along with their family members and close associates. The status is not an accusation. It signals elevated corruption and bribery risk and triggers enhanced due diligence, senior management approval, and a requirement to establish the source of wealth.

Source of wealth and source of funds are distinct, and both are asked. Source of wealth explains how the person's overall net worth was accumulated. Source of funds explains where the specific money in this relationship came from. A plausible answer to one does not answer the other.

Adverse media widens the aperture

Screening against news and public records catches what lists miss. Sanctions and PEP lists are backward-looking; an investigation or an indictment appears in the press long before any designation. Adverse media screening looks for credible reporting linking the client or its principals to financial crime, fraud, corruption, or serious regulatory breach.

It is also the noisiest control in the estate. Common names, unrelated entities, and stale articles generate enormous alert volumes, and this is where machine learning has made the most credible contribution, as Part 5 discusses.

Calibration is a documented discipline, not a dial someone turns
Figure 4 Screening volumes collapse across five stages, and the cost sits in the middle

The matching engine has a sensitivity setting, and both directions are dangerous. Set it loose and analysts drown in false positives, which raises cost and, more insidiously, trains reviewers to close alerts quickly. Set it tight and true matches are never surfaced at all, which is a control failure that produces no evidence of itself.

Supervisors therefore expect calibration to be tested rather than asserted. The standard technique is above-the-line and below-the-line testing. Above-the-line testing samples alerts that were generated and closed, to confirm the dispositions were correct. Below-the-line testing samples name pairs that fell just below the matching threshold and never generated an alert, to confirm nothing genuine was missed. The second is the one that matters, because it is the only way to observe the errors the system is not reporting.

Every threshold change is a model change. Tuning must be documented, justified, tested before deployment, and re-tested periodically, with results retained. A screening engine whose thresholds were adjusted at some point by someone for reasons nobody recorded is, from a supervisory perspective, an uncontrolled system.

The risk rating decides how deep everything else goes

Every client receives a financial-crime risk rating, and that rating governs the intensity of the whole relationship. It is calculated from the client type, its jurisdiction of domicile and operation, the products it will use, its ownership transparency, screening outcomes, and the presence of politically exposed persons. Ratings are typically low, medium, or high, sometimes with a prohibited tier above.

The rating drives three things that persist for the life of the client. It determines the depth of due diligence performed now, the frequency of periodic review, and the sensitivity of ongoing transaction monitoring. A high-risk rating is not a rejection. It is a commitment to a materially higher cost of servicing, which is why banks sometimes decline high-risk clients on economics rather than on risk.

Enhanced due diligence applies where standard measures are insufficient

Elevated risk triggers additional work rather than a different process. Enhanced due diligence typically requires establishing source of wealth and source of funds in documented detail, obtaining senior management approval to enter the relationship, more frequent review, and in some cases site visits or independent third-party reports.

Regulatory classification is collected here and consumed forever

While the client's documents are open, the bank collects the classifications that trading regulation requires. This is efficient, and it also means an error made here propagates into systems that have nothing to do with financial crime.

Regime Classification collected What it drives
MiFID II Retail, professional, or eligible counterparty Conduct protections, disclosures, reporting obligations
EMIR Financial or non-financial counterparty, above or below clearing threshold Clearing obligation, margin requirements, reporting responsibility
Dodd-Frank US person status, eligible contract participant External business conduct rules, protocol adherence
Uncleared margin rules In scope or out, by aggregate notional Initial and variation margin obligations, custody arrangements
FATCA and CRS Tax residence and status, via W-8, W-9 or self-certification Withholding and tax reporting

Tax runs in parallel, and it is more markets-specific than it looks

Tax documentation is collected here and consumed by systems the client never sees. It is easy to file mentally under administration, and it is the workstream most likely to produce a direct cash consequence for either party within weeks of go-live.

Tax residence must be certified, not assumed. Entities certify status and residence on prescribed forms, with non-United States entities using one series and United States entities another, alongside self-certification under the international exchange-of-information standard. These forms carry treaty claims where the client is entitled to reduced withholding, and a claim that is absent or defective means the full statutory rate applies.

Where the bank acts as a qualified intermediary, it takes on the withholding and reporting itself. That status simplifies life for clients and transfers responsibility to the bank, which is why the documentation supporting each account must be complete and current rather than broadly satisfactory.

The genuinely markets-specific item is withholding on dividend-equivalent payments. Rules introduced under United States tax law treat payments on certain equity derivatives that reference United States shares as if they were dividends, and subject them to withholding accordingly. The point for onboarding is that whether a future transaction falls in scope depends partly on the instrument's economics and partly on the client's certified tax status, so the determination cannot be made at trade time if the documentation was never collected. This reaches directly into equity swaps, options, and delta-one products, which is precisely the population a markets onboarding is establishing.

Transaction taxes and stamp duties add a further layer. Several jurisdictions levy a charge on transfers of shares or on transactions in specified instruments, with exemptions that depend on the client's status and role. Whether the bank collects and remits, and at what rate, is driven by fields populated during onboarding.

Tax item What it determines Where it bites
Residence and status certification Applicable withholding rate and treaty relief Over-withholding erodes client returns; under-withholding leaves the bank liable
Qualified intermediary status Whether the bank withholds and reports on the client's behalf Documentation completeness per account, examined by tax authorities
Dividend-equivalent withholding Whether equity derivative payments are treated as dividends Equity swaps, options, delta-one; cannot be determined at trade time without onboarding data
Transaction taxes and stamp duties Whether a charge applies and who remits it Settlement amounts and client billing from the first trade
Exchange-of-information reporting What is reported to which tax authority Annual reporting accuracy, driven by onboarding fields

Due diligence runs in a case management platform, and the case is the unit of record. It holds the document set, the screening alerts and their dispositions, the analyst's narrative, the risk rating calculation, and the approval trail. Regulators examine this record, so it must be append-only: dispositions can be superseded but never overwritten, and every state change carries an actor and a timestamp.

Three integrations matter more than the rest. The screening engine, called in real time at onboarding and continuously thereafter. External registry and entity data providers, which supply corporate structures and ownership. And the document repository, which must store evidence with provenance, because a passport image with no record of who collected it and when is not evidence.

The hardest technical problem here is not screening. It is entity resolution across the ownership graph, where the same underlying person appears under variant spellings across multiple registries in multiple alphabets.

Step 3 — Credit assessment prices an exposure that does not yet exist

Trading exposure is fundamentally different from lending exposure, and the difference determines how it is assessed. A loan has a known balance that declines on a schedule. A derivative portfolio has a value that moves continuously with the market, can be positive or negative, and is close to zero on the day it is struck.

The exposure is a distribution, not a number

Current exposure is what the bank would lose if the client defaulted today. It is the mark-to-market value of the portfolio, floored at zero, because a portfolio worth less than nothing to the bank represents no loss on the client's default. On day one this is usually near zero, which is precisely why it is the wrong measure to set limits against.

Potential future exposure is what the bank could lose if the client defaults later. It models how far the portfolio's value could move in the bank's favour over the life of the trades, at a high confidence level. A ten-year interest rate swap has minimal exposure today and substantial potential exposure in year four, because rates have had four years to move. Credit lines are set against this measure.

The credit valuation adjustment prices the risk into the trade. CVA is the market value of the possibility that the counterparty defaults before the portfolio matures, and it is charged to the desk executing the trade. This is the mechanism that makes counterparty credit quality a pricing input rather than an abstraction. A client with a weak credit profile receives worse pricing, not because anyone applies a penalty, but because the CVA charge on their trades is larger.

CVA has siblings, and onboarding decisions create them

The valuation adjustments are a family, and three of them are set by choices made during this process. They are the bridge between an onboarding decision and a number on a trader's screen, which is why they belong here rather than in a derivatives pricing course.

CVA is created by the credit assessment and the collateral terms. A zero threshold with daily margining collapses expected exposure and therefore collapses CVA. A five-million threshold leaves a permanent uncollateralised layer that must be priced.

MVA is created directly by the initial margin decision. Where a relationship falls in scope for regulatory initial margin, the bank must fund collateral that it posts, cannot reuse, and does not earn a market return on. The cost of funding that margin over the life of the portfolio is the margin valuation adjustment. It did not exist as a concept before the margin rules, and it is now a material component of the price of a long-dated uncleared swap. The onboarding determination of whether a client is in scope is therefore a pricing determination.

KVA is created by the netting opinion and the entity structure. Regulatory capital must be held against the exposure, and the cost of holding that capital over the life of the trades is the capital valuation adjustment. Because capital consumption depends on whether close-out netting is recognised, a clean netting opinion in the client's jurisdiction reduces KVA and a missing one inflates it. This is the concrete mechanism behind a claim made in Part 1 and developed in stage four: a legal opinion changes the price the client is quoted.

FVA completes the set by capturing the cost of funding the uncollateralised portion of a portfolio, which again traces back to the threshold agreed in the annex.

The credit line is a structure, not a single number

Approval produces a set of limits rather than one. A settlement limit covers short-dated delivery risk. A potential future exposure limit covers derivative portfolios. Tenor limits cap how far out trades may extend, because exposure and uncertainty both grow with time. Product limits restrict which instruments may be traded. And where the client is a fund, limits are set against the fund's own net assets, since that is the only pool the bank can claim against.

Credit also sets the collateral parameters that documentation will negotiate. The threshold, the minimum transfer amount, and the eligible collateral schedule are credit decisions expressed in legal language. Stage three decides them. Stage four argues about them.

Figure 5 The same two trades produce two completely different exposures, depending only on the law
Why the netting set is the unit that matters

A netting set is the group of transactions that can be offset against each other on default. Its boundaries are defined by the master agreement, and it is the true unit of credit exposure.

Suppose a fund has two swaps with the bank. One is worth plus forty million to the bank, the other minus thirty-five million. Without enforceable netting, the bank's exposure on default is forty million: it must pay the thirty-five it owes and join the queue of creditors for the forty it is owed. With enforceable netting, both terminate and net to a single claim of five million.

The reduction is a factor of eight, and it is entirely a legal outcome. Nothing about the trades changed. This is why enforceability is a capital question rather than a documentation formality, and why the netting opinion discussed in stage four has direct balance-sheet consequences.

Critically, netting sets cannot span legal entities. The Northwind bond fund and the Northwind absolute return fund each form their own netting set. A large positive exposure to one cannot be offset against a negative exposure to the other, because on default the bank would be facing two separate estates.

Step 4 — Documentation is where the commercial terms are actually made

This is the stage that determines what the relationship is worth and what it costs, and it is routinely mistaken for administration. The equivalent in a capital markets transaction is bookbuilding: the moment where price is discovered rather than assumed. Here the price takes the form of collateral terms, and it is discovered through negotiation.

The architecture of a derivatives relationship is four documents deep

Figure 6 The framework is fixed, the middle two layers are negotiated, and the bottom layer is per trade
The Master Agreement supplies the framework

The ISDA Master Agreement is a standard form that is never amended. It establishes the framework: that all transactions form a single agreement, what constitutes an event of default, what happens on termination, and how close-out amounts are calculated. Its most important provision is the single agreement concept, which is what makes close-out netting work. Because all transactions are deemed one agreement, an insolvency official cannot cherry-pick the profitable ones and disclaim the rest.

Two versions remain in wide use, from 1992 and 2002, differing principally in how close-out amounts are determined and in the length of grace periods. The 2002 form is the modern default.

The Schedule is where the negotiation happens

Everything the parties actually argue about lives in the Schedule. It selects the options the Master leaves open and adds bespoke provisions. Governing law and jurisdiction. Which entities are specified entities for cross-default. The threshold amount at which the client's other defaults trigger rights under this agreement. Whether additional termination events apply, and this is where the commercially significant clauses sit.

Additional termination events are the clauses that matter most for a fund. A net asset value decline trigger allows the bank to terminate if the fund loses a specified proportion of its value over a defined period. A key person clause triggers if named portfolio managers depart. A change of control clause triggers on a change in the fund's management. Each is a negotiation, and funds resist them because a termination right exercised in a stressed market can force liquidation at the worst possible moment.

The Credit Support Annex governs the collateral

The CSA is where credit terms become contractual, and its legal mechanics differ by jurisdiction in a way that has real consequences. The English law CSA operates by title transfer: collateral is transferred outright and the recipient owns it, subject to an obligation to return equivalent assets. The New York law CSA operates by security interest: the poster retains ownership and grants a pledge over the assets.

The difference matters on default and in insolvency. Title transfer gives the recipient clean ownership with no need to enforce a security interest, which is faster and more certain. Security interest keeps the assets in the poster's estate, which the poster prefers and which requires perfection and enforcement steps to realise.

Confirmations document individual trades

Each transaction is confirmed against the framework, specifying the economic terms while incorporating the Master by reference. This is why a trade can be agreed by phone in seconds and remain fully documented: the framework already exists, and the confirmation only supplies what is specific to that trade.

The clauses that are actually argued

Clause What it does The tension
Threshold Uncollateralised exposure permitted before collateral is called Bank wants zero, client wants operational headroom
Minimum transfer amount Smallest collateral movement, to avoid trivial transfers Bank wants low, client wants fewer operational calls
Eligible collateral What assets may be posted Bank wants cash and top-rated bonds, client wants breadth
Valuation percentage Haircut applied to posted collateral Bank wants conservative, client wants efficient
Additional termination events Bank's right to exit on defined triggers Bank wants NAV and key person triggers, client resists both
Cross-default threshold Level at which other defaults trigger rights here Both want the other's threshold lower
Valuation agent Who calculates exposure and calls margin Usually the bank, disputed by sophisticated clients
Dispute resolution Process when the parties disagree on value Client wants a binding independent process

The collateral terms are triangulated from three inputs

Figure 7 Collateral terms emerge from credit modelling, legal enforceability, and a regulatory floor

The bank's position on collateral is not a preference. It is a calculation with three components.

The internal credit assessment sets what the bank would want. Potential future exposure and CVA determine how much uncollateralised risk the bank is willing to run, which becomes its opening position on threshold.

The netting opinion determines whether the calculation is even valid. If close-out netting is not enforceable against this counterparty in this jurisdiction, the bank must measure exposure gross rather than net, which multiplies it and makes collateral terms far more demanding.

The regulatory floor overrides both where it applies. For in-scope counterparties, the uncleared margin rules mandate variation margin with zero threshold and initial margin above a specified level. Where the rules apply, there is nothing to negotiate.

Initial margin adds a second, regulated layer

Variation margin covers exposure that has already arisen. Initial margin covers exposure that might arise between a default and the close-out of the portfolio. The uncleared margin rules made initial margin mandatory for counterparties above an aggregate notional threshold, phased in over several years.

Three features make it operationally distinct from variation margin. It is calculated under a prescribed methodology, in practice the industry standard model that most participants adopted to avoid disputes. It is exchanged gross rather than net, so both parties post to each other. And it must be held by a third-party custodian in segregated accounts without rehypothecation, which requires custodial documentation and account control arrangements entirely separate from the ISDA.

A group-level threshold below which no initial margin need be exchanged softens the impact considerably, and many in-scope relationships sit below it. They must still calculate and monitor, because crossing the threshold triggers the full operational apparatus, and building that apparatus takes months.

Not every client can post collateral, and there are four alternatives

Daily margining assumes an operational capability many clients do not have. A corporate treasury with two staff cannot value a portfolio, agree a call, and move collateral before a deadline every day. Refusing to trade with such clients is one answer. In practice the market developed four substitutes, each shifting the risk somewhere else.

A guarantee moves the credit to a stronger entity. A parent company guarantee makes the group, rather than the subsidiary, the effective obligor. In the ISDA framework this party is the credit support provider, and it is named in the schedule. The bank's analysis then runs to the guarantor, which means the guarantor must itself be assessed and, frequently, onboarded.

A break clause caps the exposure by capping the time. A mandatory early termination provision requires the trade to be terminated and settled at a defined future date regardless of its remaining tenor, converting a twenty-year exposure into a five-year one. It is elegant because it reduces potential future exposure directly rather than collateralising it, and it is unpopular with clients because it reintroduces the market risk they hedged.

Third-party credit support substitutes an institution's promise for collateral, through instruments such as letters of credit. It is comparatively expensive and used where the alternatives are unavailable.

Clearing removes the question entirely by moving the exposure to a central counterparty, at the cost of the client needing clearing access and posting margin to the clearing house instead.

Each alternative is a decision made during documentation with a long tail. A guarantee creates a second onboarding. A break clause creates a diary entry that must be honoured years later by people who did not negotiate it. Neither is administrative.

Protocols let the industry repaper thousands of relationships at once

When a regulation requires the same amendment to every derivatives contract in the market, bilateral negotiation is arithmetically impossible. The industry's answer is the protocol: a standard set of amendments published centrally, which parties adhere to by lodging a signed adherence letter. Any two adhering parties have their agreements amended automatically.

This mechanism has carried the market through every major post-crisis change, from business-conduct requirements to variation margin to the replacement of interest rate benchmarks. For onboarding, the practical consequence is that checking a counterparty's adherence status is now part of the documentation workflow, and non-adherence is a gap that must be closed bilaterally at considerable cost.

Why an opinion about a distant jurisdiction moves the bank's capital

Close-out netting is a contractual right, and contractual rights are only as good as the insolvency law that will be applied to them. Many insolvency regimes give an official the power to disclaim unprofitable contracts while enforcing profitable ones. If that power survives a netting clause, the netting is worthless precisely when it is needed.

The industry addresses this by commissioning formal legal opinions on netting enforceability in each relevant jurisdiction, updated as laws change, covering the main counterparty types in each. A bank may only recognise netting for regulatory capital purposes where it holds a clean opinion covering that counterparty type in that jurisdiction.

The consequence is concrete. Two identical funds running identical portfolios, one domiciled where a clean opinion exists and one where it does not, generate materially different capital requirements for the bank. The second is more expensive to trade with, receives worse pricing, and may be declined outright. Nothing about the client's creditworthiness differs. The difference is entirely the legal system it happens to sit in.

This is also why fund domicile appears in commercial conversations that seem to have nothing to do with law.

Step 5 — Static data turns decisions into fields

Everything decided so far exists as approvals and executed contracts. None of it is yet operational. Stage five writes the decisions into the systems that will enforce them, and it is where the highest density of avoidable errors is introduced, precisely because the work looks clerical.

The record is created once and read forever

The client record consolidates the outputs of every preceding stage. Legal entity details and identifiers. The financial-crime risk rating and review date. Credit limits by type and tenor. The executed agreements and their key terms, including threshold, minimum transfer amount, and eligible collateral. Regulatory classifications. Product entitlements. Account numbers and settlement instructions.

Downstream systems bind to these fields and do not question them. The margin engine reads the threshold and calls collateral accordingly. The netting engine reads the agreement structure. The reporting engine reads the identifier and the classification. None of them re-derives anything, which is the entire point of a reference data layer and also its central danger.

Standing settlement instructions are a small field with a large blast radius

Standing settlement instructions specify where securities and cash actually go. They name the client's custodian, its account, and the correspondent chain. Every settlement instruction the bank generates for this client will use them.

They are also a standing fraud target. A convincing email requesting a change of settlement instructions, sent from a compromised or spoofed client address at a plausible moment, is one of the most effective attacks in wholesale banking. The control is callback verification to independently held contact details, never to the details in the requesting message, and it is a control that must survive commercial pressure to move quickly.

Classification flags are entered here and rarely revisited

The regulatory classifications collected in stage two become fields in stage five, and this transcription is a genuine risk point. A fund misclassified under EMIR receives wrong margin treatment. A client whose MiFID categorisation is wrong receives wrong protections and disclosures. A missing or lapsed legal entity identifier causes transaction reports to reject, and under European trading rules a client without a valid identifier cannot be traded with at all.

Reference data is the highest-leverage layer in the onboarding estate, and the design principle is single mastery. Each field has exactly one authoritative source, and every consuming system reads it rather than holding its own copy. Estates that violate this end up with a threshold of zero in the margin system and five million in the collateral system, and discover the discrepancy during a dispute.

Distribution pattern matters as much as mastery. Batch replication overnight means a limit changed at ten in the morning is not enforced until the following day. Event-driven publication, where the reference data platform emits a change event that consumers subscribe to, closes that window. Part 4 develops this.

Legal entity identifiers deserve specific mention because they expire. They must be renewed annually, and a lapsed identifier causes reporting rejections. Monitoring identifier validity is an unglamorous automated control that prevents a recurring and entirely predictable failure.

Step 6 — Connectivity makes the relationship operable

The client is approved, documented, and recorded. It still cannot trade, because nothing is connected. Stage six builds the paths that orders, margin, and settlements will travel.

Order flow requires a session and a certification

Electronic trading runs over the industry messaging protocol for order flow, and a session must be established and tested before any real order crosses it. Certification is a structured conformance exercise: the client's system and the bank's exchange test messages covering order entry, amendment, cancellation, rejection, and fill reporting, until both sides handle every case correctly.

Certification catches configuration errors that would otherwise appear as trading incidents. Wrong tag values, mishandled rejections, timestamp formats, and symbology mismatches all surface here, in an environment where the consequence is a failed test rather than a mispriced position.

Block-and-allocate is the mechanic that makes an asset manager different

A manager trading for several funds does not place one order per fund. It executes a single block in the market, because that achieves a better price and avoids competing with itself, and then allocates the fill across the funds afterwards according to a pre-agreed methodology. This is the operational counterpart of the agent-principal structure introduced at the start of this part, and it is configured during onboarding rather than invented at trade time.

The sequence has three steps and each one needs setup. The block is executed in the name of the manager acting as agent. The manager then sends an allocation instruction specifying how the block breaks down across its funds. The bank books the resulting individual transactions against each fund as principal, against that fund's own limits, netting set, and settlement instructions.

Three things must exist before this works. An account mapping that translates the manager's internal fund identifiers into the bank's client records, because the manager will not send the bank's identifiers. An allocation deadline, because the bank is carrying an unallocated block in the meantime and needs to know when it becomes a set of client positions. And a rule for what happens when an allocation arrives late, is malformed, or references a fund that is not yet trade-ready.

That last case is the one that recurs. A manager onboarded for two of its three funds executes a block and allocates a slice to the fund that has not been activated. The allocation fails, the bank holds a position it did not intend to own, and the resolution is commercial rather than systematic. Activating a manager's funds in tranches is common, and the allocation configuration must anticipate it.

Give-up arrangements are the same problem in a cleared context, where the executing broker hands the trade to a clearing broker who must accept it. An allocation and a give-up are both instructions to move a trade to its final home, and both fail in the same way when the destination is not set up.

Client money and clearing account structure are onboarding decisions

Where a client clears through the bank, how its assets are held is a choice with default-management consequences. The two broad structures are individual segregation, where the client's positions and collateral sit in an account distinct from every other client, and omnibus segregation, where they are commingled with other clients of the same clearing member.

The trade-off is protection against cost. Individual segregation gives the cleanest position on the clearing member's default, because the client's assets are identifiable and more readily transferred to another member. Omnibus is cheaper and operationally simpler, and exposes the client to the risk that shortfalls in the pooled account are shared.

Client money rules run alongside this. Where the bank holds money or assets belonging to clients, segregation, reconciliation, and record-keeping obligations attach, and the account structures that satisfy them are established during onboarding. The client must be told which structure applies and, in several regimes, must be offered the choice explicitly.

Clearing and give-up arrangements need their own setup

Where the client will clear, the clearing relationship must exist before execution can begin. Account structures are established at the clearing broker and the clearing house, give-up agreements are registered on the industry platform that administers them, and the routing that carries executed trades to the clearing broker is configured and tested.

Collateral and custody require their own plumbing

Margin cannot flow until accounts exist to hold it. Collateral accounts are opened, and where segregated initial margin applies, tri-party custodial arrangements are established with their own account control documentation. The collateral system is loaded with the CSA terms so that calls are calculated correctly from the first day.

Test transactions prove the whole chain

The final step is a small transaction taken end to end. It is executed, captured, confirmed, margined if applicable, settled, and reported. A test trade exercises every system and every interface in sequence, and it routinely finds a broken link that no component test would have caught, most often a settlement instruction that is subtly wrong or a reporting field that fails validation.

Step 7 — Activation is a status change with real force

Trade-ready is a flag in the client record, and it is load-bearing. Trading systems check it before accepting an order. The flag is not descriptive. It is the control that prevents trading with a client whose onboarding is incomplete, and it is enforced at the point of order entry rather than by policy.

Activation is granular rather than binary. A client may be trade-ready for interest rate swaps with the London entity and not yet for repo, or approved for one fund and not another. The entitlement matrix built during onboarding is what the flag actually expresses.

Step 8 — The lifecycle never ends

Onboarding is the beginning of an obligation rather than the discharge of one. The duty to know the customer is continuous, and the client record must remain accurate for as long as the relationship exists.

Periodic review re-performs due diligence on a schedule

Every client is reviewed at an interval set by its risk rating, commonly annually for high risk and every three to five years for low risk. Review re-verifies identity and ownership, refreshes screening, confirms the client's activity remains consistent with expectations, and re-rates the relationship.

Periodic review consumes more analyst capacity than new onboarding at most institutions. A bank with tens of thousands of counterparties on a rolling cycle is re-performing due diligence continuously, and the review population grows every year while the onboarding population does not.

Trigger events force review outside the cycle

Certain events require immediate review regardless of the schedule. A change of ownership or control. A screening hit arising from a list update. Adverse media. A material change in the client's activity or a shift in the risk profile of its jurisdiction. Trigger-based review is the bridge between the periodic model and the perpetual one.

Perpetual KYC replaces the calendar with events

Figure 8 The same change, detected eleven months apart or three days apart

The logic of periodic review is weak and has always been weak. A file reviewed in January and untouched until the following January is stale for eleven months, and reviewing an unchanged low-risk client on schedule spends effort where nothing has happened while a client whose ownership changed in February waits.

Perpetual KYC inverts the model. Client data is monitored continuously against external sources, and a material change triggers targeted review of what changed rather than a full re-performance. Screening runs continuously against list updates rather than at intervals. The calendar review becomes an exception rather than the mechanism.

The prerequisite is data quality that most estates do not have. Continuous monitoring requires the client record to be structured, mastered, and reliably matched to external sources. Where records are unstructured or duplicated, event-driven monitoring produces noise rather than signal, which is why the transition has been slower than the logic suggests.

Offboarding is harder than it sounds

Exiting a client is not simply deletion. Open positions must run off or be terminated, which for long-dated derivatives can take years. Collateral must be returned. Records must be retained for the statutory period, which conflicts directly with data minimisation obligations. And where the exit follows a financial-crime concern, the bank may face reporting obligations and constraints on what it may tell the client.

The residual relationship is the part people forget. A client that has been exited commercially but retains a twenty-year swap is still a counterparty. It still consumes credit limits, still requires margining, still must be screened, and still must be reviewed. Offboarding a client with long-dated positions does not end the obligation; it ends the revenue while leaving the obligation in place, which is the worst combination available and one reason banks are reluctant to exit relationships that have become uneconomic.

The internal gauntlet decides whether the client ever arrives

Figure 9 Six serial gates, each with independent authority to stop the client

Six gates stand between a sales request and a live client, and each is owned by a function with authority to refuse. Intake asks whether the bank wants this client at all. Cross-border compliance asks whether it may lawfully face them from the proposed entity. Financial crime asks whether it may have them. Credit asks what exposure it will accept. Legal asks on what enforceable terms. Operations asks whether it can actually be serviced.

The labels on the diagram refer to the three lines of defence, a control model examined properly in Part 3. In short, the first line is the business that creates the risk and manages it day to day, the second line sets standards and independently challenges the first, and the third line provides assurance to the board. Gates two through four are owned outside the business, which is the whole point of them.

None of these gates is advisory, and that is the structural point. A credit officer who declines does not make a recommendation that the front office may weigh against revenue. A financial-crime reviewer who cannot clear an alert stops the file. The gates are serial and each has veto authority, which is what makes the process a control rather than a workflow. Part 3 explains why the reporting lines behind these gates are constructed the way they are.

  1. Why does onboarding an investment manager produce more files than onboarding a corporate of similar size?

  2. What is the difference between identification and verification, and why is the distinction the substance of the control?

  3. A fund has no natural person owning above twenty-five percent. Has due diligence failed?

  4. Why is potential future exposure rather than current exposure used to set derivative credit limits?

  5. Explain how enforceable close-out netting can reduce exposure by a factor of eight without changing any trade.

  6. A CSA has a threshold of five million and a minimum transfer amount of one million. Exposure is 5.6 million and no collateral is held. Can the bank call?

  7. Why does a netting opinion in a jurisdiction the bank does not operate in affect its capital?

  8. Why are protocols used instead of bilateral amendment when regulation changes?

  9. What makes trade-ready a control rather than a description?

  10. Why has perpetual KYC been slower to adopt than its logic implies?

  11. A client negotiates a five-million threshold instead of zero. Name three prices they will pay for it.

  12. Why is below-the-line testing more informative than above-the-line testing when calibrating screening?

  13. A manager executes a block and allocates part of it to a fund that is not yet activated. What happens, and why is it foreseeable?

  14. Why does offboarding a client with long-dated derivatives produce the worst possible combination?

2

  1. What is the difference between identification and verification?

  2. A widely held fund has no person owning more than 25 %. Has due diligence failed?

  3. Why does an investment manager generate more files than a corporate of similar size?

  4. Why set derivative credit limits on potential future exposure?

3Roles, Actors, and Interactions

Four camps surround an onboarding, and only the bank must satisfy all of them

Strip the process down and four participants remain. The client, which wants access and controls most of the information the bank needs. Market infrastructure, which must be connected before anything can clear or settle. Data and document providers, which supply the evidence the bank verifies against. And regulators, which set the rules and later examine how they were applied.

Figure 10 The bank sits at the centre because it is accountable to every other party simultaneously

The bank occupies the centre for a specific reason. It is not the largest participant or the most powerful. It is the only one that must satisfy all the others at once. The client wants speed. The regulator wants rigour. The infrastructure requires precise operational conformity. The data providers supply evidence that frequently contradicts what the client has said. Every tension in onboarding is a version of the bank standing between parties whose interests do not align.

The client controls the information and pays for the delay

The client is a counterparty to the process as well as to the trades. It supplies constitutional documents, ownership evidence, authorised signatory lists, tax forms, and regulatory classifications, and it negotiates the contractual terms. Its own advisers participate directly: external counsel negotiates the documentation, and for funds the administrator and depositary supply information the manager itself may not hold.

Client responsiveness is the largest single determinant of elapsed time, which is uncomfortable for banks that measure onboarding as an internal process metric. A file waiting three weeks for a certified ownership chart is not a bank performance problem, but it appears as one in every dashboard.

Market infrastructure must be connected before anything works

Clearing houses, clearing brokers, custodians, tri-party collateral agents, and central securities depositories each require their own setup. The bank does not control any of them, and several run onboarding processes of their own. A custodian opening a segregated collateral account performs its own due diligence on the client, on its own timetable. Stage six of the flow frequently stalls on an external party that has no commercial relationship with the bank at all.

Data and document providers supply what the bank verifies against

Verification requires independent sources, and most are commercial. Corporate registry aggregators supply ownership and directorship data across jurisdictions. Sanctions and PEP list vendors consolidate designations from multiple authorities and distribute updates. Adverse media services index news and enforcement records. Identifier issuers allocate and maintain legal entity identifiers. Negotiation and signature platforms carry the documentation itself.

The quality of these feeds sets a ceiling on the quality of due diligence. A bank cannot verify ownership more accurately than its underlying registry data allows, and registry quality varies enormously by jurisdiction. This is a real limitation rather than an excuse, and it is why higher-risk jurisdictions require more human investigation.

Regulators set the rules and later grade the file

Supervisors examine onboarding files directly, and their examination is retrospective. They read what was done, when, by whom, and on what evidence, years after the fact. This is why the audit trail is not administrative overhead but the primary artefact of the process. Financial intelligence units receive suspicious activity reports, and tax authorities receive reporting driven by classifications collected during onboarding.

From To What flows Stage
Sales CLM Client request, product scope, commercial context 1
Client Bank Constitutional documents, ownership evidence, tax forms 2
Bank Data providers Screening queries on entities and individuals 2, 8
Data providers Bank Registry data, sanctions and PEP matches, adverse media 2, 8
Financial crime CLM Risk rating, approval or refusal, conditions 2
Credit risk Legal Limits, threshold, collateral parameters 3
Bank legal Client counsel Draft agreements, negotiated positions 4
Bank Custodian, tri-party agent Collateral account opening instructions 6
Bank Clearing broker, CCP Give-up registration, clearing account setup 6
Bank Client Connectivity certification, test transactions 6
CLM Trading systems Trade-ready status and entitlement matrix 7
Bank Regulators Transaction and position reporting keyed on identifiers 8
Bank Financial intelligence unit Suspicious activity reports 8

The client population is not one type but five

Client type predicts onboarding cost far better than client size does. A twenty-billion-dollar pension fund can be simpler to onboard than a two-hundred-million-dollar hedge fund, because the work tracks legal structure, product complexity, and risk profile rather than assets.

Real-money asset managers generate volume through structure

Managers running long-only mandates for institutional clients are the highest-volume onboarding population, and Northwind is the archetype. The manager itself is transparent, regulated, and easy to verify. The difficulty is entirely structural: it acts for many principals, each requiring a full file, and the population grows as the manager launches funds and wins mandates.

Their documentation position is standardised and firm. Large managers negotiate the same terms across every bank they face and resist bank-specific variations, which speeds negotiation once positions are known and makes it rigid where the bank wants something unusual. They resist additional termination events consistently, because a termination right exercised in stress could force them to liquidate positions on behalf of clients at the worst moment.

Pension funds, insurers, and sovereign wealth funds are patient and demanding

These are the longest-horizon clients and often the largest. They hedge long-dated liabilities, which means long-dated derivatives, high tenor limits, and relationships intended to last decades. Their credit quality is typically strong and their trading is predictable.

Their onboarding is slow for governance reasons rather than risk ones. Investment committees and boards approve documentation, public-sector entities may face procurement requirements, and their counsel is often external and not continuously engaged. Public pension plans routinely surface politically exposed persons among trustees, which triggers enhanced due diligence as a matter of course rather than of concern.

Hedge funds are fast, demanding, and expensive to service

Hedge funds want to trade quickly, use leverage, and trade complex products across many banks. They are the most commercially attractive clients per unit of balance sheet and the most operationally demanding. Onboarding is heavier because leverage magnifies exposure, strategies can change quickly, ownership is often less transparent, and offshore domiciles are common.

Prime brokerage compounds every one of these factors. The bank is extending secured credit against a portfolio it holds, which means concentration, liquidity, and valuation risk all sit inside a single relationship. Part 5 examines a case where a family office structured like a hedge fund caused losses in the billions, and the failure began in how the client was assessed.

Corporates hedge rather than speculate, which simplifies almost everything

A corporate treasury uses derivatives to hedge commercial exposure. Its trading is directional, low-volume, and tied to identifiable underlying business. Ownership is usually transparent, especially for listed companies, and classification is comparatively simple because most corporates fall below the thresholds that trigger the heaviest obligations.

Their weakness is capability rather than risk. Corporate treasuries often lack the legal and operational resources to negotiate a schedule or establish collateral operations, and they are the population most likely to need arrangements that avoid daily margining entirely.

Banks and financial institutions bring netting benefits and sanctions exposure

Interbank relationships are large, reciprocal, and heavily netted. Both parties are sophisticated, documentation is standard, and portfolios are large enough that netting enforceability is worth serious legal expense.

The financial-crime dimension is different in kind. A bank client may itself have thousands of customers the bank cannot see, which creates nested risk that correspondent banking rules address specifically. Jurisdiction dominates the assessment, and where a financial institution client operates in or maintains relationships with high-risk jurisdictions, the sanctions exposure is indirect and difficult to monitor.

Client type Onboarding complexity Typical risk rating Defining documents Why the bank wants or fears them
Real-money asset manager High, by multiplication Low ISDA per fund, GMRA Wants: volume, stability, low credit risk. Fears: file count per unit of revenue
Pension, insurer, SWF Medium, slow governance Low to medium ISDA, long-tenor limits Wants: size, long-dated flow, durability. Fears: elapsed time, PEP exposure
Hedge fund Very high Medium to high ISDA, CSA, PB agreement Wants: revenue per unit of balance sheet. Fears: leverage, opacity, concentration
Corporate Low to medium Low ISDA, often light CSA Wants: cross-sell, ancillary business. Fears: thin operational capability
Bank or financial institution Medium Medium to high ISDA, GMRA, netting opinions Wants: netting efficiency, scale. Fears: nested and indirect sanctions risk

Independence is the structural defence, and it is built into reporting lines

The people who want the client cannot be the people who approve it

The conflict is obvious once stated and easy to underestimate. Salespeople are compensated on revenue from clients they bring in. Those same clients must be assessed for financial-crime risk and credit appetite. If the assessors report to the people whose income depends on the answer, the assessment is not an assessment.

The defence is organisational rather than procedural. Financial-crime and credit-risk functions report through the compliance and risk hierarchies to the chief compliance officer and chief risk officer, not to the head of Global Markets. A financial-crime reviewer refusing to clear an alert cannot be overruled by the salesperson's manager, because the salesperson's manager has no authority over them. This is the whole mechanism.

The three lines of defence, defined precisely

The first line owns the risk it creates. Sales, trading, and the operational teams that support them generate the risk and are responsible for managing it day to day. Client lifecycle management sits here, which is why it coordinates but does not approve.

The second line sets standards and independently challenges the first. Compliance, financial crime, and credit risk define what good looks like, review individual decisions, and hold veto authority. They do not report through the business.

The third line provides independent assurance. Internal audit periodically tests whether the first two lines are doing what they claim, reporting to the audit committee of the board rather than to management.

The model is frequently recited and less frequently implemented. The common failure is a second line embedded in the business, sharing its objectives and reporting lines, and performing review that lacks the authority to change an outcome. Supervisors examine reporting lines and escalation records specifically because they know the recital is cheap.

What independence means in practice

Independence appears in four concrete places. Reporting lines that do not converge below executive level. Compensation that is not driven by business revenue. Explicit authority to refuse, with no commercial override available at business level. And a documented escalation path where disagreements go up the risk hierarchy rather than being settled locally.

Escalation is where the model is tested. A genuine disagreement between a relationship manager who wants a client and a reviewer who is uncomfortable is not a failure of the system. It is the system operating. The failure mode is the disagreement never being recorded, resolved informally, and leaving no trace for the examiner who arrives three years later.

This is a technology problem as much as a policy one

Independence that exists only on an organisation chart is not enforced. It becomes real when systems implement it: when a risk rating cannot be edited by someone in the front office, when an alert disposition requires a second reviewer, when an approval cannot be recorded by the person who requested it, and when every one of these actions is written to a log nobody can alter.

Five system capabilities enforce the independence model, and each corresponds to a specific failure it prevents.

Role-based access control determines who may view, edit, and approve each field. Risk ratings, screening dispositions, and credit limits must be editable only by the owning function, and the enforcement belongs in the platform rather than in policy.

Segregation of duties prevents the same identity from performing incompatible actions. The analyst who prepares a file cannot approve it, and the person who requests an exception cannot grant it.

Four-eyes approval requires a second qualified reviewer for consequential decisions, and the second identity must be genuinely distinct rather than a shared service account, which is a control that fails quietly and often.

Immutable audit logging records every decision, disposition, and override with actor, timestamp, and prior state. Records may be superseded, never overwritten, because supervisory examination reconstructs what was known at the time rather than what the record says today.

Confidentiality controls restrict suspicious activity investigations to the financial-crime function. Tipping off a client that they are under investigation is a criminal offence in most jurisdictions, so the system must make it structurally impossible for a relationship manager to see that a report has been filed.

Six internal teams run an onboarding, and the independence line runs through them

Team Division Core role Primary technology
Sales and coverage Global Markets, front office Owns the relationship, initiates the request, sets commercial context CRM
Client lifecycle management Global Markets COO Orchestrates the process, owns the client record, reports status Onboarding workflow platform
Financial crime and KYC Compliance, second line Performs due diligence, dispositions screening, sets risk rating KYC case management, screening engine
Credit risk Risk, second line Assesses exposure, sets limits and collateral parameters Credit and exposure systems
Legal and documentation Legal Drafts and negotiates agreements, confirms enforceability Contract lifecycle management, negotiation platforms
Onboarding operations Global Markets COO Sets up static data, accounts, connectivity, activation Reference data, account and connectivity platforms
Sales and coverage want the client and cannot approve them

The salesperson identifies the opportunity, initiates the request, and supplies commercial context that genuinely matters: what the client will trade, what volume is expected, and how the relationship fits the franchise. They also apply the pressure the control functions are designed to withstand, which is not a criticism of them but a description of the structure.

Client lifecycle management holds the process together

CLM is the operational centre of onboarding. It collects documents, routes the file, chases blockages, maintains the record, and communicates status to everyone. It is measured on cycle time and data quality. Because it sits in the first line, it deliberately holds no approval authority, which makes it the most influential team in the process and the least powerful.

Financial crime and KYC hold the hardest veto

Analysts perform due diligence, investigate ownership, disposition screening alerts, and set the risk rating. Their authority is absolute within its scope: a file they will not clear does not proceed, and no commercial argument reaches them through a reporting line. They are also the function under the most sustained pressure, which is precisely why their independence is structural rather than cultural.

Credit risk decides what the bank is exposed to

Credit officers model potential future exposure, set limits by product and tenor, and specify the collateral parameters that documentation will negotiate. They face a version of the same pressure, and hold the same independence.

Legal and documentation convert decisions into enforceable terms

Negotiators draft and argue the schedule and the annex, and the legal function confirms enforceability, including netting analysis and reliance on jurisdictional opinions. They translate a credit decision into contractual language that will hold up in an insolvency proceeding years later, in a jurisdiction they may never visit.

Onboarding operations make it real

Operations creates accounts, populates static data, loads limits and collateral terms, establishes connectivity, runs certification and test transactions, and sets the trade-ready flag. Their work involves the least judgement and the highest density of consequential detail, and a settlement instruction entered wrongly here will fail trades long after everyone has forgotten the onboarding.

Two specialist functions join at defined points

The six teams above run the process end to end. Two others enter at specific moments and leave, which is why they are easy to omit from an organisation chart and expensive to omit from a plan.

Cross-border compliance owns the permissibility determination. It sits within legal or compliance depending on the institution, and it answers the second gate: may this booking entity provide these products to this client in this territory. It is consulted at scoping and again whenever scope changes, and a scope change late in the process reopens its analysis. Teams that treat it as a formality discover that it is the one function whose negative answer cannot be mitigated by any commercial or contractual arrangement.

Tax operations owns certification and withholding status. It validates the tax forms collected during due diligence, determines the applicable withholding treatment, and maintains the expiry monitoring that prevents a lapsed certification from silently reverting a client to the full statutory rate. Its work is invisible until the first payment is withheld incorrectly, at which point it becomes extremely visible.

Neither function has veto authority in the way credit and financial crime do, with one exception: cross-border compliance effectively does, because an activity the bank is not licensed to conduct is not a risk to be priced but a line that cannot be crossed.

Every stage maps to a team and a system

Stage Lead team Key system
1. Intake and scoping Sales with CLM CRM and onboarding workflow
2. KYC and financial crime Financial crime KYC case management, screening engine
3. Credit assessment Credit risk Exposure modelling and limits systems
4. Legal documentation Legal and documentation Contract lifecycle management
5. Static and reference data Onboarding operations Client and reference data platform
6. Connectivity and testing Onboarding operations with IT Connectivity, collateral, clearing platforms
7. Activation CLM Client record and entitlement matrix
8. Ongoing lifecycle CLM with financial crime Perpetual monitoring and review workflow

  1. Why does the bank sit at the centre of the actor map rather than the client?

  2. Why can a small hedge fund be a heavier onboarding than a large pension fund?

  3. How does client type change the prioritisation decision, and why is revenue per file a misleading measure?

  4. Why does client lifecycle management deliberately hold no approval authority?

  5. What are the four concrete markers of genuine independence?

  6. Why must a relationship manager be structurally prevented from seeing that a suspicious activity report has been filed?

  7. Why is an unrecorded, informally resolved disagreement between a reviewer and the front office worse than an escalated one?

3

  1. Why does client lifecycle management hold no approval authority?

  2. Why can a small hedge fund be heavier to onboard than a large pension fund?

  3. Why is revenue per file a misleading way to prioritise the queue?

  4. Why does the bank, not the client, sit at the centre of the actor map?

4Systems, Data, and IT

Ask the clean question first: what functions does any onboarding need?

Name the functions before naming the vendors. Every bank's onboarding estate looks different and every one performs the same eight functions, because the functions follow from the obligations rather than from any architectural choice. Starting from the function makes the vendor landscape legible instead of bewildering.

Figure 11 Eight functional blocks, and five capabilities that cut across all of them
Functional block What it does What it produces
Client portal and outreach Requests, receives, and tracks client-supplied documents and data Evidence with provenance, and a visible status for the client
Intake and CRM Captures the request, scopes entities and products, resolves against existing clients Prospect record and scope
KYC case management Holds the due diligence case, evidence, narrative, and approvals Verified client file and risk rating
Screening engine Matches names against sanctions, PEP, and adverse media data Alerts and dispositions
Contract and document management Drafts, negotiates, executes, and stores agreements Executed agreements and extracted terms
Credit limits and collateral Models exposure, holds limits, calculates margin Limit set and collateral parameters
Client data and accounts Masters the client record, accounts, and settlement instructions The authoritative client record
Connectivity and enablement Establishes order, clearing, collateral, and settlement paths Working operational connections

Five capabilities cut across every block

Entity resolution determines whether two records describe the same legal person, across internal systems and external sources. It is the hardest technical problem in the estate and is discussed at length below.

Workflow and case orchestration routes files between functions, enforces the sequence, tracks status, and escalates delays. This is what makes onboarding a process rather than a set of queues.

Immutable audit records every decision with actor, timestamp, and prior state, because supervisory examination reconstructs what was known at the time.

Data privacy governs the personal data of beneficial owners, directors, and signatories, which is collected in volume and is subject to protection rules that pull against retention obligations.

Perpetual monitoring watches client data and external sources continuously and raises events when something material changes.

Real vendors fill each block, and where banks build tells you where the edge is

Functional block Typical provider category Buy or build
Client portal and outreach Secure document exchange, identity verification services Buy
Intake and CRM Enterprise CRM platforms Buy
KYC case management Specialist client lifecycle platforms, low-code workflow suites Buy, configure heavily
Screening engine Watchlist data vendors and screening software Buy
Registry and ownership data Corporate data aggregators, identifier foundations Buy
Contract management and negotiation Industry negotiation platforms, contract lifecycle suites, e-signature Buy
Credit exposure modelling Trading and risk platforms, plus proprietary models Mixed, models built
Limits and risk appetite logic Proprietary Build
Client risk rating methodology Proprietary Build
Reference data mastering Enterprise data management platforms Buy, model built
Entitlement and trade-ready enforcement Proprietary Build
Collateral and margin Specialist collateral platforms, industry reconciliation utilities Buy
Regulatory reporting Trade repositories, approved reporting mechanisms Buy

The client-facing layer is the one most banks under-build

Part 3 identifies client responsiveness as the largest single determinant of elapsed time. It is worth pausing on how strange it is, then, that the client-facing layer receives the least investment of any block in the estate. Banks build sophisticated internal workflow and then request documents by email.

A serious client layer does four things. It states precisely what is required, per entity, with examples, rather than sending a generic checklist that a fund administrator must interpret. It accepts documents securely, with the provenance metadata that turns an uploaded file into evidence. It shows the client what has been received, what is outstanding, and what is blocking activation, which removes an enormous volume of status chasing in both directions. And it remembers, so that the same document is not requested again at periodic review three years later.

Automated verification belongs here too. Identity verification for directors and signatories, registry lookups that pre-populate ownership data, and validation that a document is legible and current before a human ever opens it. Each removes a round trip, and round trips are measured in weeks when the client is a fund administrator in another time zone.

The reason this layer is under-built is organisational rather than technical. Internal workflow is owned and measured by the operations function, while the client experience sits between sales and operations and belongs to neither. The metric that would justify the investment, elapsed time excluding bank-controlled work, is precisely the one most institutions do not measure.

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

Banks buy where the requirement is identical for everyone and build where it encodes their own risk appetite. Screening against a sanctions list is the same problem for every institution, and the data comes from the same authorities, so there is no advantage in building it and considerable disadvantage in maintaining it. Deciding what risk rating a Luxembourg fund with a politically exposed director should receive is a statement of the bank's own appetite, and no vendor can supply it.

The same logic explains the two most consistently proprietary components. The client risk rating methodology and the entitlement matrix that enforces trade-ready status are built almost everywhere, because both encode decisions the bank must own and defend to a supervisor. A bank that outsources its risk rating methodology has outsourced the thing it will be examined on.

Data lineage stitches the blocks into one machine

This is the most important idea in the technology picture, and it is worth stating plainly. Onboarding does not produce a decision. It produces a record, and that record is read by systems that will never re-derive anything in it.

Figure 12 The client record is created once and consumed by every downstream system
Stage System Data object created
1. Intake CRM and workflow Prospect record: entities, products, scope
2. Due diligence KYC case management, screening Verified identity, ownership graph, dispositions, risk rating
3. Credit Exposure and limits systems Limit set, netting set definition, collateral parameters
4. Documentation Contract lifecycle management Executed agreements and extracted key terms
5. Static data Reference data platform Authoritative client record: identifiers, classifications, accounts, instructions
6. Connectivity Connectivity, collateral, clearing platforms Sessions, accounts, custodial arrangements
7. Activation Client record Trade-ready flag and entitlement matrix
8. Live operation Margin, settlement, reporting engines Margin calls, settlement instructions, regulatory reports

Follow one field and the cascade becomes concrete

Take the EMIR classification of the Northwind absolute return fund. It is established in stage two from documents the client supplies, transcribed into the client record in stage five, and then read by four separate systems that never question it.

The margin system reads it to decide whether initial margin applies. The clearing determination reads it to decide whether trades must be centrally cleared. The reporting engine reads it to determine which party reports and what fields the report carries. And the collateral system reads the associated terms to calculate calls.

Now suppose it is wrong. A fund classified as a non-financial counterparty when it is in fact a financial counterparty will have trades that should have been cleared executed bilaterally, margin that should have been exchanged not exchanged, and regulatory reports submitted with incorrect fields. None of these systems will detect the error, because each is functioning exactly as designed on the data it was given. The error surfaces months later, usually through a regulatory reporting reconciliation or a supervisory query, and remediating it means unwinding every consequence back to the original transcription.

This is the cascade principle, and it is why onboarding data quality is a first-order risk rather than an operational nicety. An error at the front does not stay at the front. It propagates into every system that trusts the record, and the further downstream it is discovered, the more expensive it is to unwind.

Anatomy of the client record

The client record is the atom of this entire module. Everything the process does is either populating it or reading it. It is worth knowing what it actually contains.

Field group Examples What it drives downstream
Identity Legal name, former names, registered address, incorporation number, legal form Screening, verification, contract naming
Identifiers LEI, internal client ID, BIC, tax identifiers Reporting, settlement, matching
Hierarchy Parent, ultimate parent, agent and principal links, fund to manager mapping Exposure aggregation, netting set boundaries
Ownership and control Beneficial owners, directors, authorised signatories Screening scope, instruction authority
Financial crime Risk rating, review due date, EDD status, PEP flags Review frequency, monitoring sensitivity
Regulatory classification MiFID category, EMIR classification, UMR scope, tax status Margin, clearing, reporting, conduct obligations
Credit Limits by product and tenor, netting set identifier Pre-trade limit checks, exposure aggregation
Documentation Agreement type and version, governing law, threshold, MTA, eligible collateral, protocol adherence Margin calculation, close-out treatment
Operational Accounts, settlement instructions, custodian, clearing broker, session identifiers Settlement, clearing, connectivity
Status Trade-ready flag, entitlements by product and booking entity Order acceptance at point of entry

Entity resolution is the hard problem, and it never fully goes away

The question sounds trivial and is not: are these two records the same legal person? The same fund appears in the bank's systems as an abbreviated trading name, in a corporate registry under its full legal name, in a sanctions list transliterated from another alphabet, and in a legacy platform under a name it stopped using after a merger. None of the strings match, and all four refer to the same entity.

Getting this wrong fails in both directions, and both are expensive. Failing to match creates duplicate client records, which fragment exposure so that a client sits within limits on each record while breaching in aggregate, and which cause the same client to be onboarded and reviewed several times. Matching wrongly merges two distinct entities, which is worse: it can attach one client's credit line, classification, or clean screening result to another.

The technique that works is graph-based rather than string-based. Modern entity resolution builds a network from identifiers, addresses, directors, ownership links, and prior names, and scores the likelihood that two nodes are the same entity using the structure of the network rather than the similarity of the text. This is one of the places where machine learning has produced genuine improvement rather than marketing, because the problem is pattern recognition over messy multilingual data, which is exactly what these methods are good at.

Identifiers: nothing flows until everything is named the same way

Names are unreliable keys and identifiers exist to replace them. Four matter in this process.

The legal entity identifier names the counterparty

A twenty-character code identifying a legal entity that participates in financial transactions, issued under a global federated system and carrying reference data on the entity and its ultimate parent. It is the central identifier of this module, because it is what regulatory reporting is keyed on.

It expires, which is the operational point most often missed. Identifiers must be renewed annually, and a lapsed identifier causes transaction reports to reject. Under European trading rules the position is stricter still: without a valid identifier for the client, the trade cannot be executed. Monitoring validity across a client population is a simple automated control that prevents an entirely predictable recurring failure.

The ISIN names the instrument

A twelve-character code identifying a specific security, used in settlement and reporting. Onboarding does not create ISINs, but the client record must be able to reference the instruments a client is entitled to trade.

The BIC names the institution in the settlement chain

An eight or eleven-character code identifying a financial institution, used to route settlement messages. It appears in standing settlement instructions and is how the custodian and correspondent chain is expressed.

Internal and national identifiers name what the others do not

Every bank maintains its own client identifier, because internal systems need a stable key that does not depend on external issuance. Separately, transaction reporting requires natural persons to be identified by prescribed national identifiers, which brings personal data into the reporting chain and connects directly to the privacy obligations discussed below.

Identifier What it names Where it bites
LEI The legal entity Reporting rejects and trading is blocked if invalid or lapsed
ISIN The instrument Settlement and reporting mismatches
BIC The institution in the settlement chain Failed settlement, misrouted instructions
Internal client ID The client inside the bank Duplicate records, fragmented exposure
National identifier The natural person Transaction reporting rejections, privacy obligations

How the systems talk: batch, API, and events

Batch is the legacy default

Files are assembled and transferred on a schedule, typically overnight. Batch is simple, well understood, and adequate where the data changes slowly. Its weakness is latency, and in this estate latency has consequences: a credit limit reduced at ten in the morning is not enforced by downstream systems until the following day, leaving a window in which trades are accepted against a limit that no longer exists.

APIs make data available on demand

A system requests exactly what it needs at the moment it needs it. Screening calls are the natural example, because a name must be checked at the moment a file is assessed rather than in a nightly cycle. APIs remove staleness for the requester but do not by themselves inform anyone that something changed.

Event streaming pushes changes the instant they happen

The publishing system emits an event when data changes, and every subscriber receives it immediately. A risk rating change, a limit reduction, a sanctions hit, or a lapsed identifier is published once and consumed by everyone who cares, within seconds.

The migration from batch to event streaming is usually justified as a speed improvement, and that undersells it considerably. The deeper change is that it removes the need for reconciliation.

Under batch replication, every consuming system holds its own copy of the client record, and those copies drift. The bank therefore builds reconciliation processes to detect divergence, exception queues to investigate it, and operations teams to resolve it. That entire apparatus exists because the copies can disagree.

Under event streaming with a single publisher, there is one authoritative record and consumers subscribe to changes. Copies cannot silently drift, because there is no independent path by which they could. The reconciliation processes, the exception queues, and much of the operational headcount attached to them become unnecessary rather than faster.

This is the distinction between an improvement and a transformation, and Part 5 applies the same test to every trend it examines. Speeding up a reconciliation is an improvement. Removing the reason it exists is a transformation.

Some records must be impossible to change

Immutable audit is the primary artefact of the process

Supervisory examination is retrospective and reconstructive. An examiner asks what the bank knew at the time, what it decided, who decided it, and on what evidence. Answering requires a record of the past state, not the current one.

The requirement is therefore append-only storage for compliance-critical records. Screening dispositions, risk ratings, approvals, exceptions, and overrides may be superseded by later entries but never overwritten, and every entry carries an actor and a timestamp. A system that updates a risk rating in place, leaving no trace of the previous value or who changed it, cannot answer the examiner's question and has destroyed evidence rather than maintained a record.

The risk rating engine is a rules engine with a conscience

Risk rating is computed rather than judged, at least initially. A rules engine combines client type, jurisdictions, products, ownership transparency, screening outcomes, and PEP exposure into a rating, applying the bank's own weightings.

Two properties make it different from an ordinary rules engine. It must be explainable, because a supervisor will ask why a specific client received a specific rating and the answer must be reconstructable rather than asserted. And it must permit documented override, because judgement sometimes departs from the model, but every override must be attributed, justified, and reviewable. An override facility with no record is an unlogged bypass of the bank's stated risk appetite.

The messages that carry a live client

Onboarding ends in connectivity, and connectivity means messaging standards. Three layers matter, and they operate on completely different timescales.

FIX carries the order conversation

The Financial Information eXchange protocol is the language of electronic order flow. It carries orders, amendments, cancellations, executions, and allocations between the client's systems and the bank's, in structured messages exchanged over persistent sessions in milliseconds.

Onboarding's involvement is establishing and certifying the session, which is the conformance testing described in stage six.

It is also where allocation messaging is configured, and for an agent client this is more than a detail. The protocol carries dedicated message types for the block-and-allocate sequence described in Part 2: the execution of the block, then a separate allocation instruction breaking it into individual client positions. That instruction identifies each destination fund by an account code the manager maintains, which means the bank must hold a mapping from the manager's codes to its own client records.

That mapping is an onboarding artefact and a recurring source of failure. It is created when the funds are set up, and it breaks whenever the manager adds a fund, renames an account, or restructures internally without telling anyone. A mapping that silently goes stale produces allocation rejections at the point where a position must be booked, under time pressure, on a trade that has already been executed in the market.

SWIFT carries settlement and account instructions

Post-trade messaging runs on a different network with different economics. Settlement instructions, confirmations, and account transfers move as structured messages between institutions in minutes to hours rather than milliseconds, because settlement operates on a timescale of days.

The industry is migrating from the legacy message types to a richer standard, which carries structured data in an extensible format and permits far more information per message. For onboarding the practical consequence is that standing settlement instructions and account data must be captured with the granularity the new format expects, and estates built around the older format's constraints require remediation.

The regulatory pipes carry reports keyed on the client record

Transaction and position reporting flows to trade repositories and approved reporting mechanisms, on a next-day cycle. These reports are keyed on the legal entity identifier and populated from classification fields established during onboarding, which is the point at which the whole chain closes: a field entered by an operations analyst in stage five becomes a regulatory submission with the bank's name on it.

Layer Standard Carries Timescale
Order flow FIX Orders, executions, allocations Milliseconds
Post-trade settlement SWIFT, migrating to ISO 20022 Settlement instructions, confirmations Minutes to hours
Collateral and margin Industry margin platforms and messaging Margin calls, disputes, IM reconciliation Intraday
Regulatory reporting Repository and reporting formats Transaction and position reports Next day

Resilience: screening cannot fail, and neither can go-live

Some moments have no second chance

Two moments in this process are genuinely unforgiving. The first is continuous sanctions screening, because a designation published while the screening platform is unavailable does not wait. Transactions executed against a newly designated party during an outage are breaches, and the defence that the system was down is not a defence.

The second is the client's first day of trading. Go-live concentrates every newly configured component into a single sequence, and a failure there is visible to the client at the exact moment the relationship is being formed.

Resilience is built from four things, and rehearsal is what makes them real

Concept What it guarantees Why it matters here
Redundancy A spare component exists and is ready Instance or hardware failure does not stop screening
Automatic failover Traffic switches without human intervention Screening cannot pause while someone is paged at three in the morning
Disaster recovery Service restored at a second site within a defined objective A regional outage does not suspend an entire client population
Business continuity People and process continue, not only systems A manual fallback exists while the platform is unavailable
Rehearsal The plan is known to work rather than believed to An untested recovery plan is a document, not a control

Five regulatory regimes read as system specifications

Each of these is usually summarised as law. Read them instead as requirements documents, because that is how they land on the estate.

Anti-money-laundering rules specify the file

They mandate what must be collected, verified, and retained, on a risk-sensitive basis, with prescribed enhanced measures above defined thresholds. As a specification they require a structured case model, evidence storage with provenance, a risk rating capability with defined inputs, review scheduling driven by that rating, and retention for a statutory period. They also require the whole thing to be reconstructable years later, which is the immutability requirement.

Sanctions regimes specify the gate

They prohibit dealing with designated parties, on a strict-liability basis. As a specification they require screening of every entity and associated individual against multiple authorities' lists, at onboarding and continuously thereafter, with fuzzy matching to survive transliteration and name variation, alert workflow with recorded dispositions, and list-update handling measured in hours rather than days. The strict-liability character is what makes availability an architectural requirement.

MiFID II specifies the classification and the reporting

It requires clients to be categorised, with the category driving protections and disclosures, and requires detailed transaction reporting. As a specification it requires the categorisation to be a controlled field with an audit trail, identifiers to be validated before trading, personal identifiers for natural persons, and reporting extracts assembled from onboarding data with completeness and accuracy obligations attached.

The derivatives rulebook specifies the margin and the clearing

Post-crisis derivatives regulation across jurisdictions requires counterparty classification, clearing of standardised products, margin exchange for uncleared products, and reporting. As a specification it requires classification fields per counterparty, a clearing determination capability, initial and variation margin calculation with an agreed methodology, segregated custody for initial margin, and portfolio reconciliation with counterparties. It is the single largest driver of field count in the client record.

Data protection specifies the limits on everything above

The beneficial owners, directors, and signatories collected during onboarding are natural persons with rights over their data. As a specification it requires a lawful basis for processing, minimisation, accuracy, defined retention, and the ability to respond to access requests.

It also creates a genuine tension that has no clean resolution. Anti-money-laundering law requires retention for a statutory period; data protection requires deletion when no longer necessary. Sanctions screening requires processing personal data at scale; minimisation requires collecting no more than needed. These are reconciled through documented lawful bases and defined retention schedules rather than by choosing a side, and the reconciliation must be defensible to two different regulators with different priorities.

Two further regimes regulate the systems rather than the client

The five regimes above tell the bank what it must know about a client. Two more tell it what its systems must be able to do, and they are the ones most often missing from a discussion that stays close to the business.

Risk data aggregation principles specify the lineage itself

Supervisory principles on risk data aggregation and reporting were issued after the financial crisis for a blunt reason: banks could not answer basic questions about their own exposures quickly enough. They require that risk data be accurate, complete, timely, and adaptable, that its lineage be documented, and that aggregation be possible across the group rather than only within silos.

Read as a specification, this is the formal requirement behind everything argued earlier in this part. Single mastery of each field. Documented lineage from origination to report. The ability to aggregate exposure to one client across entities and systems, which is exactly what duplicate client records defeat. A bank that cannot demonstrate where a classification field came from and which systems consume it is not merely operating a messy estate; it is non-compliant with an explicit principle.

The principles apply to risk data, and client reference data is risk data, because exposure aggregation depends entirely on entity identity and hierarchy. This is the regulatory hook that converts entity resolution from an engineering preference into an obligation.

Operational resilience rules regulate the estate and its suppliers

Newer operational resilience regulation, most prominently the European digital operational resilience framework, imposes direct requirements on information and communications technology risk. It requires identification of critical functions, defined tolerances for disruption, incident reporting within tight deadlines, and, crucially, scenario testing including threat-led penetration testing for significant institutions.

Its most consequential feature for this estate is third-party oversight. It brings critical technology providers into scope, requires contractual rights of audit and defined exit strategies, and establishes direct oversight of designated critical providers.

That matters here because onboarding depends on a handful of external suppliers. A small number of vendors supply the consolidated sanctions and PEP data used across the industry. A small number supply corporate registry and ownership data. A small number operate the negotiation and identifier infrastructure. Concentration on this scale means an outage or a data quality failure at one provider is not an isolated incident; it is a simultaneous degradation of the same control at many institutions at once.

The practical obligations are unglamorous and specific. Know which providers are critical to which functions. Hold contractual audit rights. Have a documented, tested plan for continuing to screen if a primary data feed becomes unavailable, which usually means a secondary source rather than a hope. And be able to report an incident within the deadline, which requires knowing that it happened.

  1. Why do banks buy screening engines but build risk rating methodologies?

  2. Trace what happens when a fund's EMIR classification is entered incorrectly.

  3. Why is failing to match two records for the same entity dangerous, and why is wrongly merging two entities worse?

  4. What does event streaming remove that batch replication requires, and why is that a transformation rather than an improvement?

  5. What distinguishes FIX from settlement messaging, and why must onboarding establish both?

  6. Why is redundancy insufficient for sanctions screening resilience?

  7. Why must compliance records be append-only rather than updatable?

  8. Why does a lapsed legal entity identifier stop trading rather than merely causing a reporting exception?

  9. How are the conflicting demands of anti-money-laundering retention and data protection minimisation reconciled?

  10. Why do risk data aggregation principles turn entity resolution from an engineering preference into an obligation?

  11. Why does vendor concentration matter more in this estate than in most?

  12. Why can a genuinely global client record often not exist as a single physical store?

  13. Where is the largest share of onboarding elapsed time lost, and why is that block the least invested?

  14. An allocation instruction from a manager is rejected because it references an account code the bank does not recognise. What broke, and when was it created?

4

  1. Why do banks buy screening engines but build risk rating methodologies?

  2. A fund's EMIR classification is entered wrongly. What follows?

  3. Which entity-matching error is worse?

  4. What does event streaming from a single source remove?

5Risks, Pain Points, and Transformation

A MECE map of where onboarding risk lives

Before the war stories, a clean taxonomy. Onboarding risk is usually discussed as a single undifferentiated worry about compliance, which makes it impossible to allocate ownership or design controls. Six types cover the field, and each has a different owner and a different defence.

Figure 13 Plotting the six risks by controllability and severity shows where effort belongs and where judgement does
Risk type What it is Who owns it Main defence
Financial crime and sanctions Accepting a client the bank should have refused, or missing a designation Compliance, second line Verification depth, screening coverage, continuous monitoring
Credit and counterparty Exposure larger or more volatile than assessed, or collateral that fails Credit risk, second line Exposure modelling, limits, collateral terms, review
Legal and enforceability Agreements that do not hold in insolvency, netting that fails Legal Netting opinions, contractual discipline, jurisdiction control
Operational and static data Wrong fields, wrong instructions, wrong entitlements, vendor failure Onboarding operations Single mastery, four-eyes, test transactions, validation
Independence and conflict Commercial pressure changing a control outcome Compliance, second line Reporting lines, segregation of duties, logged overrides
Reputational Association with a client that damages the franchise Senior management Intake gate, escalation, appetite statements

Third-party failure belongs inside operational risk rather than beside it. A sanctions data feed that degrades, a registry provider that goes down, or an identifier service that becomes unavailable produces exactly the same outcome as an internal system failure: the control does not operate. Part 4 explains why the concentration of these providers makes the exposure larger than it appears, and why operational resilience regulation now reaches them directly. Keeping it inside the operational category preserves the taxonomy's discipline while making clear that the defence includes secondary sources, audit rights, and tested contingency rather than only internal controls.

How to tell a good onboarding function from a fast one

Cycle time is the metric everyone uses and the easiest to improve dishonestly. A file can always be closed faster by accepting weaker evidence, and the cost of doing so appears in a different quarter under a different heading. A serious measurement set therefore pairs speed with quality.

Metric What it measures Why it matters
Cycle time to trade-ready Elapsed days from intake to activation Time to revenue, and the client's experience of the bank
Bank-controlled time Elapsed time excluding waiting on the client Separates process performance from client responsiveness
First-time-right rate Share of files completed without rework The honest counterweight to cycle time
Quality assurance pass rate Share of files passing independent sampling Detects deterioration before a supervisor does
Backlog and file ageing Volume and age distribution of open files Ageing distribution predicts breaches better than average age
Periodic review currency Share of clients reviewed within due date The most common supervisory finding in this area
Screening false positive rate Alerts dispositioned as non-matches Drives cost, and tuning it downward can hide true hits
Remediation volume Clients reopened for previously accepted files The accumulated interest on past speed

Three real failures cover the three ways onboarding breaks

Nothing teaches this subject faster than watching it fail at scale. The three cases below are not chosen for size, though they are among the largest. They are chosen because each fails at a different point in the process described in Part 2.

Danske Bank Estonia — a client base that should never have been accepted

What happened

Danske Bank acquired a Finnish bank in 2007, and with it an Estonian branch that ran a substantial non-resident customer portfolio. Between 2007 and 2015, roughly two hundred billion euros flowed through that portfolio, largely from Russia and other former Soviet states, through customers that included large numbers of opaque shell structures. Danske's own investigation, published in 2018, found that a significant proportion of those customers and transactions were suspicious.

The consequences were severe. The chief executive resigned in 2018. The Estonian regulator ordered the bank to cease operations in the country in 2019. In December 2022 Danske pleaded guilty in the United States to conspiracy to commit bank fraud and agreed to a resolution of approximately two billion dollars.

Root cause

The failure was at the front of the process, not the end of it. These clients were onboarded. The due diligence performed on them was inadequate in ways that a functioning process would have caught: ownership structures were opaque, the stated business rationale for the accounts was implausible, and the customer profile did not match the transaction behaviour.

One technical fact deserves particular attention in this module. The Estonian branch operated on a separate IT platform that was never integrated with the group's systems. Group-level screening, monitoring, and reporting did not cover it. The controls the bank believed it had were not applied to the population that most needed them, and nobody at the centre could see the client records well enough to know.

The lesson

A control that does not cover a population does not exist for that population. Acquisitions and regional platforms are where this failure lives, because integration is expensive and deferred, and the deferral is invisible until someone looks. Coverage is a question that must be asked explicitly, per population and per system, rather than assumed from a group policy statement.

BNP Paribas — controls that existed and were deliberately circumvented

What happened

Between roughly 2004 and 2012, BNP Paribas processed US dollar transactions on behalf of entities in countries subject to United States sanctions, including Sudan, Iran, and Cuba. The conduct was not accidental. Transactions were routed through intermediary institutions and identifying information was removed from payment messages so that the transactions would not be flagged by sanctions filters in the United States.

In 2014 the bank pleaded guilty and agreed to pay approximately 8.9 billion dollars, the largest sanctions penalty imposed on a financial institution, together with a temporary suspension of certain US dollar clearing activities.

Root cause

The screening controls worked. They were routed around. This is a different failure from Danske. The technology functioned as designed, and the failure was that people who understood exactly how it functioned removed the data it depended on. Stripping identifying information from a message defeats even a perfectly tuned screening engine, because the engine can only match what it receives.

The independence failure sits underneath the conduct failure. Concerns were raised internally and did not stop the business. Where commercial importance is allowed to determine whether a control outcome is respected, the control has been converted into an opinion, and Part 3 describes the reporting-line architecture that exists specifically to prevent this.

The lesson

A control is only as strong as the integrity of the data reaching it, and the independence of the people operating it. Screening effectiveness is usually measured on matching accuracy, which assumes complete and honest input. The harder assurance question is whether anything upstream can modify or suppress what the control sees, and whether the people who could do so face any consequence.

Archegos — the right questions asked of the wrong risk

What happened

Archegos Capital Management was a family office that built extremely large, highly concentrated equity positions through total return swaps with several prime brokers simultaneously. Because the positions were held synthetically through derivatives, and because each bank saw only its own exposure, no counterparty had visibility of the aggregate.

In March 2021 the underlying positions fell, margin calls could not be met, and the resulting forced unwind produced losses across the prime brokerage industry of well over ten billion dollars. Credit Suisse alone lost approximately 5.5 billion. Its independent report described a fundamental failure of management and controls. Nomura, Morgan Stanley, UBS and others took losses running from hundreds of millions into the billions. The principal was subsequently convicted of fraud and racketeering in 2024.

Root cause

This failure sits squarely in stages three and four of the process. The client was onboarded, assessed, and documented. The problem was that the assessment underestimated the risk and the documentation locked in terms that could not respond to it.

Three specific onboarding failures compound in this case. The margin terms agreed at documentation were static rather than dynamic, so they did not tighten as the portfolio became more concentrated and more leveraged. The client-type assessment treated a family office as a lower-risk category than the hedge fund it functionally was, despite a strategy that was unambiguously a leveraged concentrated equity fund. And the principal's history included a prior regulatory resolution for insider dealing, which was publicly available and is exactly the sort of finding adverse-media screening exists to surface.

Ongoing monitoring failed to correct any of it. Exposure grew substantially over months, and where risk functions did identify the concentration and sought to increase margin, the changes were not enforced with the urgency the position warranted.

The lesson

Onboarding sets terms that must survive conditions nobody has yet seen, and static terms are a bet that the client will not change. For clients whose risk profile can move quickly, the terms themselves must be dynamic, and periodic review must have genuine authority to retighten them. The related lesson is structural and uncomfortable: no single bank could see Archegos in aggregate, because each held a legitimate view of its own exposure only. Some risks are invisible to a correctly performed bilateral assessment, and recognising that limitation is part of doing it correctly.

Case Failure type Where it maps Lesson
Danske Bank Estonia Wrong clients accepted Steps 1 and 2, intake and due diligence A control that does not cover a population does not exist for it
BNP Paribas Controls circumvented Step 2, screening, and the independence model A control is only as strong as its input integrity and its operators' independence
Archegos Risk misjudged, terms too loose Steps 3, 4 and 8, credit, documentation, review Static terms are a bet the client will not change, and some risk is invisible bilaterally

Generative AI arrives in the KYC file, with governance close behind

Large language models are unusually well matched to parts of this process and unusually badly matched to others, and the boundary is sharp enough to state precisely.

Where they genuinely help

Document extraction and summarisation is the strongest use. An onboarding file contains constitutional documents, fund prospectuses, ownership charts, and regulatory filings, frequently across languages. Extracting structured fields from unstructured documents is exactly what these models do well, and it attacks a task that consumes a large share of analyst time and creates no judgement value.

Adverse media triage is the second. Screening a name can return hundreds of articles, most irrelevant. A model can summarise, cluster, and assess relevance far faster than an analyst reading sequentially, presenting a shortlist with reasoning attached rather than a raw result set.

Contract review during negotiation is the third. Comparing a client's markup against the bank's standard positions, identifying deviations, and flagging clauses that depart from the playbook is a well-defined comparison task on text, and it compresses the second effort peak directly.

Analyst support is the fourth and least discussed. Onboarding policy is voluminous and jurisdiction-specific, and a retrieval-based assistant that answers procedural questions with citations reduces both errors and the time senior staff spend answering the same questions.

The risks are specific and the governance is not optional

Confidentiality is the first and most immediate. KYC files contain personal data on beneficial owners and directors, commercially sensitive information about client strategy, and material the client supplied in confidence. Sending any of it to a model outside the bank's control is a data protection and confidentiality breach before it is anything else. This constraint drives architecture toward private deployment and rules out a range of otherwise attractive tools.

Hallucination is dangerous here in a particular way. A model that fabricates an ownership percentage or asserts that a screening match is unrelated produces output that looks exactly like correct output. In a file that a supervisor will examine years later, a plausible fabrication is worse than a gap, because a gap is visible and a fabrication is not.

Explainability collides directly with model risk governance. Financial-crime decisions must be reconstructable and defensible. A model that cannot explain why it reached a conclusion cannot support a regulatory decision, which is why the credible deployments keep the model on the drafting and triage side of the line and leave determination to a human.

Bias is the least discussed and most consequential. A model trained on historical risk ratings will reproduce whatever patterns those ratings contained, including any that functioned as proxies for nationality or jurisdiction in ways that would not survive scrutiny if stated explicitly. Testing for this requires deliberate effort, because the model will not volunteer it.

Use of generative AI The gain The governing risk
Document extraction and summarisation Removes the largest non-judgement time cost in the file Hallucinated fields entering the client record
Adverse media triage Turns hundreds of articles into a reasoned shortlist Missed true positives, and confidentiality of queries
Contract clause comparison Compresses negotiation and surfaces deviations Misread clauses reaching an executed agreement
Analyst policy assistant Fewer procedural errors, less senior time on repeat questions Confidently wrong procedural guidance
Draft narrative for KYC files Faster file completion Explainability, and narrative that outruns the evidence

  1. Why is reputational risk owned at senior level and defended at the intake gate rather than through process controls?

  2. A function halves cycle time while remediation volume doubles. What has actually happened?

  3. Distinguish the Danske and BNP Paribas failures precisely. — Danske is a due diligence failure: clients were accepted that should have been refused, on a branch platform group controls never covered, so the controls did not exist for that population. BNP Paribas is an integrity and independence failure: the controls worked but identifying data was stripped from messages so the screening engine could not match, and internal concerns did not stop the business.

  4. Why does the Archegos case belong to documentation and credit rather than to KYC?

  5. What did Archegos demonstrate about the limits of a correctly performed bilateral assessment?

  6. Which of the five trends changes the operating model rather than the tooling, and why?

  7. Apply the transformation filter to digitising an onboarding form. — It changes appearance. The same analyst performs the same assessment on the same evidence through a different interface. Cycle time may improve marginally, but no cost pool is removed and no risk is reduced, so it fails the test.

  8. Why must generative AI stay on the drafting and triage side of the line in financial crime work?

  9. Why is a hallucinated field in a KYC file worse than a missing one?

5

  1. Why is reputational risk defended at the intake gate?

  2. Cycle time halves while remediation doubles. What has happened?

  3. Why does Archegos belong to documentation and credit rather than KYC?

  4. What kind of failure was Danske?

Module Summary — The Six Ideas That Matter Most

The client, not the trade, defines the platform

Onboarding effort tracks the duration and volatility of the exposure a relationship creates, not its size or revenue. A large equity order settling in two days is a simpler onboarding than a small thirty-year swap, because the swap binds the parties while the market moves underneath them. This is why the seven routes order as they do, why bilateral derivatives is the reference case, and why an asset manager acting for four principals generates four times the work for one commercial conversation. Design the function around exposure-time and the workload becomes predictable. Design it around client size and it will be wrong in both directions. And before any of that arithmetic matters, one question outranks it: whether the bank may lawfully face this entity from this booking entity at all, which is the cheapest gate to pass and the most expensive to skip.

Data cascades, so errors are never contained

The process produces one object that everything downstream trusts unconditionally. The client record carries identity, identifiers, hierarchy, ownership, risk rating, regulatory classifications, credit limits, documentation terms, settlement instructions, and entitlements, in several hundred fields. Margin engines read the threshold without re-deriving it. Reporting engines read the classification without re-verifying it. Order entry reads the trade-ready flag without questioning it. A single wrong classification becomes a wrong margin call, a missed clearing obligation, and a rejected regulatory report, and none of the affected systems will detect it because each is working correctly on the data it was given. This is why stage five, which looks clerical, carries more downstream risk than stage three, which looks analytical.

Documentation is where the terms are actually made

The schedule and the collateral annex are the bookbuilding of this process. A threshold argued over during negotiation determines how much exposure runs uncollateralised for years. A minimum transfer amount determines when the bank may call at all. A haircut schedule determines how much collateral is really worth. An additional termination event determines whether the bank can exit when a fund deteriorates, and the Archegos case shows what static terms cost when the client changes faster than the contract anticipated. These terms are triangulated from credit modelling, netting enforceability, and a regulatory floor, and the enforceability input means a legal opinion about a jurisdiction the bank may never operate in changes the capital it holds against the relationship. They also reach the trading floor directly: threshold drives CVA and FVA, initial margin scope drives MVA, and netting enforceability drives KVA, so terms settled by lawyers months earlier sit inside every price the client is quoted.

Independence is architecture, not policy

The functions that can stop a client do not report to the people who want the client, and that single structural fact is the main defence. Financial crime and credit risk sit in the second line with veto authority, compensation not driven by business revenue, and escalation running up the risk hierarchy. The model becomes real only when systems enforce it: role-based access so ratings cannot be edited by the front office, segregation of duties so preparation and approval are separate identities, four-eyes on consequential decisions, immutable logging so examination can reconstruct what was known, and confidentiality controls that make tipping off structurally impossible. BNP Paribas is what it costs when controls exist and independence does not.

The economics of the queue quietly distort behaviour

Onboarding capacity is finite and therefore rationed, and rationing is where commercial pressure enters a control process. Prioritising the queue by expected revenue and strategic importance is legitimate and necessary. Allowing commercial value to influence the standard applied to a file is neither, and the distinction is easy to state and genuinely difficult to hold, because the same senior people set both. Ranking purely on revenue per file also misleads on its own terms, since it favours clients who generate revenue quickly over clients whose relationships are worth more across a decade. What a client costs to run, which follows from client type, is the number that separates a profitable relationship from a busy one.

Real transformation moves cost or risk

Apply one filter to everything proposed in this space. Digitising a form changes appearance. Removing the reconciliation between two copies of the client record changes economics, because the team, the queue, and the error class disappear together. Replacing scheduled review with event-driven monitoring changes both, which is why perpetual KYC is the strongest of the five trends. Generative AI passes the filter where it extracts, triages, and drafts, and fails it where it determines, because a supervisory explanation that terminates in a model output is not an explanation. The human in the loop is a control, not a limitation, and the deployments that have delivered real gains are the ones that made the analyst's work easier without making their judgement optional.

Process Cheat Sheet — The Whole Onboarding on One Page

A single reference for the entire markets counterparty onboarding flow, from first request to perpetual monitoring.

Stage Timing Core activity Key output Lead and systems
1. Intake and scoping 1-2 wks Define entities, products, booking entities; confirm permissibility Scope and permissibility matrix Sales with CLM; CRM, workflow
2. KYC and financial crime 3-8 wks Verify identity, ownership, control; screen; risk rate; collect tax status Approved file and risk rating Financial crime; case management, screening
3. Credit assessment 2-4 wks Model PFE, set limits, fix collateral parameters Limit set and CSA parameters Credit risk; exposure and limits systems
4. Legal documentation 4-16 wks Negotiate schedule and annex, confirm enforceability Executed agreements Legal; contract management, negotiation platforms
5. Static and reference data 1-2 wks Populate client record, accounts, settlement instructions Authoritative client record Onboarding ops; reference data platform
6. Connectivity and testing 2-4 wks Certify sessions, map allocations, open clearing and collateral accounts Working operational paths Onboarding ops with IT; connectivity, collateral
7. Activation Days Set entitlements and trade-ready status Live client CLM; client record
8. Ongoing lifecycle Perpetual Review, monitor, refresh, remediate, offboard Current accurate record CLM with financial crime; monitoring platforms

Key numbers to remember

Metric Rule of thumb
Execution-only cash client, end to end 2-4 weeks
Bilateral derivatives client, end to end 3-6 months
Prime brokerage client, end to end 4-8 months
Beneficial ownership threshold Commonly 25%, treated as a floor not a ceiling
Legal entity identifier 20 characters, renewed annually or reporting rejects
Tax certification validity Commonly three years, then full statutory withholding applies
Periodic review cycle Annual high risk, 3-5 years low risk
Internal teams touching a complex file 6-8, plus cross-border compliance and tax
Serial gates that can stop a client 6, each with veto authority
Fields in a complete client record Several hundred
Netting effect, illustrative Positions of +40 and -35 reduce exposure from 40 to 5
Typical CSA threshold for a fund Zero, so all exposure is collateralised
Margin call test Call only if exposure less threshold less collateral held exceeds the MTA
Price of a higher threshold Larger CVA and FVA, plus higher capital consumption
What creates MVA and KVA Initial margin scope, and netting enforceability
Largest single driver of elapsed time Waiting on the client
Largest ongoing cost pool Periodic review, not new onboarding

The six gates, in order

Gate Question it asks Owner
1. Intake and scoping Do we want this client at all? 1LOD, sales and CLM
2. Cross-border permissibility May we lawfully face them from this entity? Legal and compliance
3. Financial crime sign-off May we have them? 2LOD, compliance
4. Credit approval What exposure will we accept? 2LOD, credit risk
5. Legal documentation executed On what enforceable terms? Legal and documentation
6. Activation and trade-ready Can we actually service them? 1LOD, onboarding operations

The one-sentence version

A bank establishes that it may lawfully face a counterparty and who that counterparty really is, decides how much exposure it will accept and on what enforceable terms, writes every one of those decisions into a client record that margin, clearing, reporting, and order-entry systems will trust without ever re-deriving it, connects the operational paths, flags the client trade-ready, and then monitors all of it for as long as the relationship lasts.

Sources and further reading

The three failure cases in Part 5 are matters of public record rather than industry anecdote. This section identifies where the underlying material sits, and which bodies publish the standards the module relies on.

The three cases

Case Primary public material
Danske Bank Estonia The bank's own commissioned investigation into the Estonian non-resident portfolio, published September 2018; the Estonian regulator's 2019 precept requiring the bank to cease operations in the country; the United States Department of Justice resolution of December 2022
BNP Paribas The United States Department of Justice and New York Department of Financial Services resolutions of 2014, including the agreed statement of facts describing the routing and message-stripping practices
Archegos The independent external report commissioned by the Credit Suisse board into its Archegos exposure, published July 2021; the subsequent United States criminal proceedings concluded in 2024

Industry standards and documentation

Derivatives documentation. ISDA publishes the Master Agreement, the Credit Support Annex and its jurisdictional variants, the protocols referred to throughout Part 2, and commissions the netting and collateral enforceability opinions that Part 2 explains are a capital question rather than a legal formality.

Financing and lending documentation. ICMA publishes the Global Master Repurchase Agreement. ISLA publishes the Global Master Securities Lending Agreement. FIA publishes the give-up templates and operates the platform on which they are administered.

Identifiers. GLEIF operates the global legal entity identifier system and publishes the reference data behind it.

Regulation referred to in this module

Derivatives. EMIR and its subsequent revisions in Europe; Title VII of the Dodd-Frank Act in the United States; and the margin requirements for non-centrally cleared derivatives issued jointly by the Basel Committee and IOSCO, which are the source of the uncleared margin rules.

Conduct, categorisation and reporting. MiFID II and MiFIR.

Financial crime and sanctions. The FATF Recommendations, which set the international standard; the successive European anti-money-laundering directives; and the sanctions lists maintained by the relevant national and supranational authorities.

Tax. FATCA and the OECD Common Reporting Standard; the United States rules on dividend-equivalent payments discussed in Part 2.

The systems themselves. The Basel Committee principles for effective risk data aggregation and risk reporting; and the European digital operational resilience framework, including its provisions on critical third-party providers.

Capstone Exercise — Scope an Onboarding You Have Not Seen

The self-check boxes test recall. This tests judgement. Work through the brief before reading the solution, and write your answers down, because the value is in committing to a position rather than recognising one.

The brief

A salesperson forwards the following request. It is deliberately written the way these requests actually arrive, which is to say incompletely.

"Meridian Partners want to start trading with us. They're a credit fund manager, about four billion under management, based in Singapore. They want rates swaps and repo, and they've asked about clearing some of it. Their head of trading used to work here. They'd like to be live before the end of the quarter because they have a hedging programme starting. Can you get this moving?"

Preliminary enquiries establish the following. Meridian Partners Pte Ltd is the Singapore-incorporated manager, regulated locally. It manages two vehicles that would trade: Meridian Credit Opportunities Fund, a Cayman Islands exempted company whose investors are largely European institutions, and a segregated mandate for a Japanese insurance company, where the insurer is the principal. The manager is owned sixty percent by its two founders and forty percent by a private equity holding company incorporated in Mauritius, whose own ownership is not immediately apparent. The bank would book through its London and Singapore entities.

Your task

Answer these six questions before reading on. Write the answers out.

1. How many onboarding files does this request generate, and what is each one?

2. Which gate is most likely to stop or delay this client, and why?

3. Which single fact in the brief should worry you most, and what would you do about it?

4. Which product creates the heaviest documentation burden, and what would you expect to negotiate hardest?

5. Where will the elapsed time actually go, and is the end-of-quarter deadline realistic?

6. Name three fields established during this onboarding that would cause a downstream failure if entered incorrectly.

Worked solution

1. Four files, not one

The manager and three principals. Meridian Partners Pte Ltd is assessed because it instructs the bank, even though it carries no trading liability. The Credit Opportunities Fund, the Japanese insurer, and, if the mandate is structured through a separate vehicle rather than the insurer directly, that vehicle as well. On the facts given, the insurer is the principal, so the count is four entities in total.

Each principal needs its own everything. Verification, identifier, credit line sized against its own assets, documentation, classification, and netting set. The Cayman fund and the Japanese insurer cannot be netted against each other under any circumstances.

2. Cross-border permissibility, gate two

Three jurisdictions are in play on the client side and two on the bank side, producing a permissibility matrix rather than a yes or no. Can the London entity face a Cayman fund managed from Singapore for uncleared swaps. Can either entity face a Japanese insurer, which is a regulated domestic institution in a jurisdiction with its own rules on foreign counterparties. Which entity faces the manager for the purpose of receiving instructions.

Some cells will close. The likely outcome is that Singapore books the Asian relationships and London books the European-investor fund, but that is a determination for cross-border compliance and it must be made before due diligence begins rather than after.

3. The Mauritius holding company

Forty percent of the manager is held through a vehicle whose own ownership is not apparent, and that is the single most important fact in the brief. Forty percent is above any beneficial ownership threshold, so the chain must be followed through the Mauritius entity to the natural persons behind it. If that ownership proves genuinely opaque, or if the structure appears designed to obscure rather than to serve a tax or commercial purpose, the file does not proceed on the strength of the manager's local regulatory status.

What you do about it is specific. Request certified ownership evidence for the Mauritius entity and every layer above it until natural persons are reached. Screen each of them. Establish source of wealth if any prove to be politically exposed. Expect this to take weeks and to be the critical path, and set that expectation with the salesperson immediately rather than when it becomes a delay.

The head of trading being a former employee is not relevant to any of this, and is worth naming only because it is the kind of detail that generates informal pressure to move quickly.

4. Uncleared rates swaps, and the argument will be about termination rights

Repo under a master agreement with annexes is substantial but comparatively standardised. The clearing element is lighter still, because the exposure moves to the clearing house.

The uncleared swaps carry the full stack and, for a credit fund, the hardest negotiation is over additional termination events. A net asset value decline trigger is exactly what the bank will want against a credit strategy that can lose value quickly, and exactly what the fund will resist, because a termination right exercised in a stressed credit market forces liquidation at the worst possible moment. Expect this to be the clause that takes the longest.

Expect initial margin scope to be tested too. Whether the relationship falls within the uncleared margin rules determines whether segregated custody arrangements must be built, which adds months if they must.

5. The deadline is not realistic, and the reason matters

Six to nine months is the honest estimate, against a request for roughly one quarter.

The time goes in three places. Ownership resolution through the Mauritius chain, which is investigative and depends on evidence the client must obtain. Cross-border determination across three client jurisdictions and two booking entities, which is sequential with everything else. And documentation, where the termination-event negotiation will run through the fund's counsel and probably its investment committee.

The right response is not to promise the quarter and miss it. It is to state the estimate with the reasons attached, identify what the client can do to compress it, which is principally supplying ownership evidence immediately, and if a hedging programme genuinely cannot wait, ask whether a narrower initial scope could be activated first. Clearing-only or repo-only activation is materially faster than the full set, and Part 2 explains why.

6. Three fields with downstream consequences

Many candidates are defensible. Three obvious ones follow.

The regulatory classification of the Cayman fund. It drives clearing determination, margin treatment, and which party reports. Wrong here and trades are executed bilaterally that should have been cleared, margin is not exchanged that should have been, and reports are submitted with incorrect fields, none of which any system will flag.

The netting set boundary between the fund and the insurer. If the hierarchy wrongly links them, exposure is aggregated across two entities that cannot legally be netted, and the bank believes it is safer than it is.

The standing settlement instructions for the Japanese mandate. A cross-border settlement chain with a local custodian is precisely where an incorrect field produces failed settlement, and it will not be discovered until the first trade attempts to settle unless a test transaction is run.