Answer first: A cloud exit plan is credible only after a test proves that priority data can be exported and reconciled, applications and dependencies can be rebuilt securely, commercial and access rights survive stress, service can remain within tolerance, and the firm understands the people, time and transition bill. A backup alone is not portability.

Key takeaways

  • Separate planned migration, provider disruption and provider insolvency scenarios.
  • Inventory data, managed services, identities, keys and operational knowledge that resist export.
  • Define success in time, functionality, cost, reconciliation and impact tolerance.
  • Rebuild security and operations as well as application availability.

Decision framework

Decision area Evidence to require Risk it controls
Data Priority export completes and reconciles Locked, incomplete or stale records
Application Critical function runs in target environment Managed-service dependency blocks rebuild
Security Identity, keys, logging and patching work Recovered but unsafe service
Operations Runbooks, staff and support can operate it One-off technical demo
Commercial Rights, fees, assistance and transition period are usable Exit blocked during dispute

A fintech can leave a cloud provider credibly when it can move the necessary data, rebuild required controls and keep important customer services operating within an acceptable transition. Downloading a database is only one component. Identity, security configuration, integrations and the people who operate them may be much harder to transfer.

The question is not whether a startup should avoid managed services. Those services can save valuable engineering time. The question is whether management understands the dependency it is accepting and can afford a realistic exit if circumstances change.

Separate planned migration from emergency exit

A negotiated move over twelve months differs from losing access to an essential service during stress. Build separate scenarios. The first can optimise cost; the second prioritises continuity, records and control.

PRA SS2/21 addresses outsourcing and third-party risk management for firms within its scope, including exit planning. 2 A fintech supplier should understand the buyer's requirements without claiming that every expectation applies directly to every software business.

Choose the services that matter most. A nonessential dashboard and a customer-balance system do not need identical exit designs. Proportionality should be a documented decision based on impact, rather than a reason never to test anything.

Inventory what cannot be exported cleanly

List application data, transaction history, encryption keys, audit records, access policies, monitoring configuration and infrastructure definitions. Record formats, volumes, retention needs and who can initiate export.

NHS England's cloud-exit guidance highlights migration and the limited portability of identity and security-policy metadata. 1 It is a healthcare framework, not fintech regulation, but the technical lesson travels: moving records does not automatically reproduce permissions and safeguards.

Test one representative workload in another environment. Include a historical query, a new transaction, access revocation and a recovery operation. A successful file transfer proves little if the business cannot interpret or use the transferred information.

Calculate the transition bill

Include data transfer, parallel running, new engineering, specialist support, testing, contract termination and staff time. Add performance differences that may increase the cost of the replacement platform.

An illustrative exit costing £150,000 may be acceptable for a company with sufficient cash and six months to move. The same estimate is inadequate if it excludes £80,000 of parallel infrastructure or assumes engineers who are already committed to a regulatory deadline.

Model a stressed version with less time and fewer available staff. A reserve estimate is meaningful only if it connects to the scenario and can actually be funded when required.

Check commercial rights before the dispute

Review access after termination, export assistance, deletion, fees, subcontractor dependencies and audit evidence. Distinguish the contractual right to obtain data from a technically workable process to retrieve it.

Understand what the provider controls and what your company controls. If the only administrator account belongs to an external contractor, the exit problem may begin before the cloud contract does.

For customer-owned data, consider the obligations you have promised downstream. A startup cannot offer a bank a three-day transition if its own critical provider offers no corresponding assistance or capability.

Rebuild security, not just availability

The replacement environment needs appropriate identity, logging, encryption, monitoring and incident processes. The NCSC's cloud principles provide a useful assessment structure.

3 Apply it to the actual target configuration rather than assuming another major provider is automatically equivalent.

Reconcile financial records after migration. Compare counts, totals, states and selected transaction histories. Preserve a defensible cut-over point and a rollback decision. A migration that loses a small number of critical events can create a large customer problem.

Keep test evidence and record what failed. The first exercise is allowed to expose gaps; a plan that repeatedly reports success without resolving them is less useful.

Choose what must be portable

Portability does not require rebuilding every feature in the lowest common denominator.

Decide which records, interfaces and controls are indispensable, then identify where deliberate lock-in produces enough value to justify its cost.

For a hypothetical early-stage analytics product, a proprietary query engine may accelerate delivery. The company can accept that dependency while preserving customer data in an interpretable format and documenting the work needed to replace the engine.

Record that judgement in the product architecture and commercial plan. Otherwise a reasonable initial trade-off can become an accidental permanent commitment that nobody revisits.

Give the test a measurable finish line

Specify the dataset, operations and service level the exercise must demonstrate. Include a transaction created before migration and queried afterwards, a permission removed during the move and a correction made in the target environment.

Measure elapsed time, staff effort, transfer cost and unresolved functionality. Do not report success simply because the destination database contains the expected number of rows.

Compare the result with the scenario's available time and money. If the test takes three weeks under ideal conditions but the stressed exit assumes five days, the gap needs an action or an explicit acceptance decision.

Keep the test repeatable. A one-off migration performed by the only engineer who understands the system may demonstrate personal expertise rather than organisational capability. Documentation and a second operator help establish whether the plan survives a real change in circumstances.

What should a founder decide?

Accept some lock-in deliberately when the managed capability creates enough value. Reduce it around indispensable data, difficult-to-reconstruct controls and customer-critical services.

Review the decision as volume and obligations grow.

Accept managed-service dependence where its benefits justify the exposure. Preserve practical choices around indispensable records and controls. A tested exit makes the cost of changing provider visible before the company is forced to pay it.

A practical 90-day implementation plan

  1. Select one service and define planned and stressed exit assumptions.
  2. Export priority data and verify completeness, schema, lineage and restore time.
  3. Rebuild the minimum viable service with fresh identity, keys, logging and operational ownership.
  4. Measure elapsed time, service gaps, cost and residual lock-in; update contract and architecture decisions.

Frequently asked questions

Is a backup a cloud exit plan?

No. A backup may protect data, but exit also requires usable formats, dependencies, identity, keys, infrastructure, people, contracts and a target operating environment.

Must every workload be portable?

Not necessarily. Apply proportionality: identify what supports important services, what can be replaced, what can be retired and where lock-in is an accepted board-level risk.

What is a successful exit test?

Success is measured against pre-agreed time, data integrity, functionality, security, operational and cost criteria under a stated scenario.

Reporting by Shoreditch Talk