Negotiate the Exit Before You Sign: Termination, Data Egress and Model Portability in Fraud Technology Contracts

Every fraud and financial crime technology contract ends. It ends in renewal, in replacement, in a vendor acquisition that changes the roadmap, or occasionally in a failure that forces a scramble. Yet in the selection processes TrustSphere runs and observes, exit is the least negotiated section of the contract and often the least read. Teams that will argue for weeks over a licence fee breakpoint sign exit schedules that give them thirty days of reasonable assistance at unspecified rates, and discover five years later what that sentence costs.

Our house view is that exit terms are a day-one procurement issue, not a lawyer's tidy-up at signature. The leverage to obtain a real exit package exists at exactly one moment: before the contract is signed, while there are still two vendors at the table. After signature the vendor is a monopolist over your own history, your tuned rules, your trained models and your case data, and every one of those assets becomes a renewal negotiation hostage. We would go further: an institution's ability to leave a detection vendor in an orderly way is itself a financial crime control, because a forced, badly supported migration is a period of degraded detection, and criminals do not pause for your transition programme.

This is also no longer just good practice. For EU financial entities, DORA requires documented exit strategies for ICT services supporting critical or important functions, with Article 28 expecting firms to be able to exit without disrupting business activities or compliance with regulatory requirements. The EBA Guidelines on outsourcing arrangements have required documented, tested exit strategies for critical or important outsourced functions since 2019, and in the UK the PRA's supervisory statement SS2/21 sets equivalent expectations on business continuity and exit plans. For most fraud and AML platforms sitting in the transaction path, arguing that the function is not critical or important is a conversation we would not want to have with a supervisor.

Where procurement gets exit wrong

The first failure is sequencing. Exit terms arrive in the draft master services agreement after vendor selection, when the business has announced the winner internally, the programme plan assumes a start date, and nobody will reopen commercial points. The vendor's lawyers know this, which is why the standard exit schedule is standard.

The second failure is treating exit as data return. Getting a database dump back is the trivial part. The assets that actually determine whether you can leave are the tuned detection configuration: rule logic and thresholds refined over years of your own loss experience, machine learning models trained wholly or partly on your data, list and watchlist configurations, investigator case histories with their attachments and audit trails, and the mapping documentation that explains what any of it means. If those are contractually the vendor's confidential information, or returned as unusable exports, you do not have an exit plan. You have a rebuild plan with a deadline.

The third failure is pricing. Exit assistance clauses without a rate card, a defined service description and a committed duration are invitations to invoice. A vendor supporting your migration to its competitor has no commercial incentive to be cheap, fast or thorough, and goodwill is not a contractual term.

A composite from TrustSphere's recent engagements shows how the failures compound. A mid-sized European bank decided to replace a transaction monitoring platform after seven years. The contract returned its data as a proprietary-format archive, licensed the rule configuration as vendor confidential information, and promised reasonable migration assistance at then-current professional services rates. The eventual exit took nineteen months against a planned nine. The bank paid for a parallel-running period nearly twice as long as budgeted, rebuilt some two hundred tuned scenarios from investigator memory and sampled alert output because the logic itself could not be exported, and accepted a known detection gap during cutover that had to be disclosed to its supervisor. Every one of those costs was determined by sentences agreed seven years earlier, in a schedule nobody at the selection stage had read against a migration scenario.

What to negotiate, specifically, before signature

Run exit as a scored requirement inside the RFP itself. Ask every shortlisted vendor the same questions in writing and score the answers alongside detection capability, because the answers differ far more than the marketing does.

• Data egress: a complete export of all client data, case records, audit trails and attachments, in documented open formats with schemas supplied, available on demand throughout the term and not only at termination, with the first full export exercised during implementation as an acceptance test.

• Rule and configuration portability: an explicit term that detection rules, thresholds, scenario logic and tuning developed on your data are your property or at minimum perpetually licensed to you, delivered in human-readable form on request.

• Model portability and explanation: where models are trained on your data, the contract states what you receive on exit, which may be the model artefact, the training specification, or at minimum full feature definitions and performance documentation sufficient to reproduce behaviour, and says so now rather than leaving it to a future negotiation.

• Exit assistance: a defined service, a named maximum rate, a minimum committed period of cooperation that outlives termination for cause in both directions, and continuity of service levels during the transition window.

Two further points separate serious buyers. First, notice and wind-down asymmetry: the vendor's right to terminate or suspend should carry a longer notice period and stronger continuity obligations than yours, because the operational harm is asymmetric. Suspension rights that can switch off screening over a billing dispute have no place in a financial crime contract. Second, test the exit plan rather than filing it. The EBA guidelines expect exit plans for critical functions to be sufficiently tested; in practice that means periodically exercising the data export, timing it, and checking someone other than the vendor can read the output.

Evaluating what comes back

Vendor responses on exit sort into three recognisable grades. The best vendors answer with specifics: named formats, a sample export, a published exit methodology, reference clients who have actually left. Treat a vendor that can describe its own funerals calmly as lower risk everywhere else in the relationship too. The middle grade answers in principle but resists committed rates and durations; that is negotiable, and the RFP stage is where to negotiate it. The worst grade claims exit terms are not standard, cites intellectual property in everything including your own tuning, or prices egress as a premium service. Score that honestly as delivery risk, because a vendor that fights you over a hypothetical exit will fight harder over a real one.

TrustSphere's engagement data suggests the market rewards the effort. In the selection processes we have supported over the past two years, buyers who scored exit provisions as a weighted RFP criterion obtained materially stronger packages, committed egress formats, capped assistance rates and tuning ownership, in the majority of cases, at no measurable premium on licence fees. Vendors concede exit terms at selection time because they are competing; the same request a year into the term is a change request, priced like one.

Both sides of the table, and the second line

For the buyer, the discipline is to staff exit negotiation properly: procurement cannot do it alone, because the questions about model artefacts, export schemas and detection continuity are technical. Put a data architect and a financial crime operations lead in the room when the exit schedule is drafted.

For the vendor, clean exit terms are increasingly a sales asset rather than a concession. Buyers under DORA and SS2/21 must evidence exitability to their supervisors; a vendor with a documented exit methodology, open export formats and a reference list of orderly departures removes a compliance blocker from its own deal. Vendors who understand this are starting to lead with it.

For the second line and the outsourcing risk function that must sign off, the test is simple: could this institution leave this provider within the notice period without a detection outage, and is there evidence, not assertion, behind the answer? If the exit plan has never produced a tested export and the register entry says exit risk: low, the register is describing hope.

Suggested Next Steps

• Add exit, egress and portability as a weighted, scored section of every fraud and financial crime technology RFP, with written vendor answers that survive into the contract as warranted statements.

• Exercise a full data export during implementation as a formal acceptance criterion, and repeat the test at least annually, timing it and validating that the output is readable without the vendor's tools.

• Inventory your current contracts against DORA Article 28 and EBA outsourcing expectations, flagging any critical detection service with undefined exit assistance, uncapped rates or vendor-owned tuning, and prioritise remediation at the next renewal.

• Write the exit strategy for your highest-dependency detection platform now, naming the realistic replacement options, the migration duration evidenced by the tested export, and the interim detection controls that would run during a transition.

Sources: EU Council and Commission, DORA, Regulation (EU) 2022/2554, EBA, PRA, TrustSphere Risk Index, April 2026.

TrustSphere helps financial institutions design and deploy intelligent fraud and financial crime detection solutions. Visit www.trustsphere.ai