Answer first: A payment outage is not recovered when an API returns 200 responses. Recovery requires a known financial position: accepted instructions are classified, duplicates are controlled, external settlement is reconciled, balances and customer messages agree, manual adjustments are governed, and the important business service is restored within its impact tolerance without leaving latent harm.

Key takeaways

  • Define the customer outcome, not the component, as the important business service.
  • Freeze and classify the financial state before replaying uncertain work.
  • Test boundary cases such as timeout after external acceptance.
  • Reconcile service status, ledger state, settlement and customer communication.

Decision framework

Decision area Evidence to require Risk it controls
Intake Know which instructions were accepted or rejected Lost or duplicate payment
Ledger Balances, holds and journals form a known state Incorrect customer balance
External rail Settlement/acquirer files reconcile to internal records Unseen financial exposure
Customer Status and support scripts reflect actual state Misleading reassurance
Recovery Replay and manual adjustments are idempotent and approved Second incident

A payment service has not recovered merely because customers can log in again. Recovery also requires reliable balances, known transaction states and a controlled way to resolve payments interrupted by the incident. The app, internal ledger and external payment provider must tell a consistent financial story.

For a London fintech, this is both a customer-outcome question and a commercial one. An apparently short outage can create days of manual reconciliation, duplicate refunds and support contacts if state recovery was never tested.

Define the service customers depend on

The FCA's operational-resilience framework applies to firms in scope and centres on important business services and impact tolerances. Firms in scope had until 31 March 2025 to be able to remain within their impact tolerances. 1 Other technology businesses can use service-based recovery testing without assuming the same regulatory perimeter.

A useful exercise begins with a customer outcome: making a payment, receiving funds or accessing an accurate balance. It then identifies the people, systems, data and external services needed to deliver it.

Restoring a database or restarting an API is an internal action. The service definition should explain when customers can safely resume the activity and what residual limitations remain.

Freeze a known financial position

During an incident, preserve transaction identifiers, event timestamps, provider responses and the last known reconciliation point. Avoid uncontrolled manual changes that remove the ability to reconstruct the sequence.

Classify payments into confirmed success, confirmed failure, pending and unknown. Unknown is a real state. Automatically retrying an unknown payment can produce a duplicate if the first instruction succeeded but its response was lost.

The recovery plan should specify how staff query the provider, match external records and authorise corrections. Give high-risk actions separate review. The person solving the technical outage should not have to invent financial adjustment rules under pressure.

Test the awkward boundary conditions

Simulate a timeout after the provider accepts an instruction but before the app receives confirmation. Deliver an event twice. Deliver it late. Restore a backup whose timestamp precedes some externally settled payments.

For each case ask whether money moves once, the ledger records it once and the customer receives an accurate status. Include fees, foreign exchange, refunds and reversals where they exist in the product.

These tests reveal a different class of risk from server availability. A system can be responsive while every subsequent balance inherits an earlier error. Performance monitoring alone cannot detect that failure reliably.

Measure harm beyond minutes offline

Track affected customers, delayed obligations, unresolved payment value and the age of uncertain transactions. A short disruption at payroll time can matter more than a longer interruption of a rarely used feature.

The Bank of England describes resilience as a financial-sector concern that includes firms and their dependencies. 2 Translate that broad objective into service-specific measures rather than adopting one recovery-time number across the entire business.

An illustrative incident affecting 200 transfers worth £400,000 is not equivalent to one affecting 200 account-profile updates. Both need technical recovery; only one immediately creates uncertainty over money movement. Scenario design should reflect that difference.

Give support a truthful status

Prepare messages for unavailable service, degraded service, pending payment and confirmed recovery. Avoid promising that balances are correct before reconciliation has finished.

Support needs a case view that reflects verified transaction states and prevents staff from initiating duplicate refunds. Customers should not have to retell the incident to each department or receive contradictory advice from the app and call centre.

The NCSC's cloud-security principles provide a framework for assessing supporting controls and assurance. 3 They do not replace product-specific recovery tests. Ask what evidence shows the complete payment service can operate under the chosen scenario.

Reconcile the customer message with the financial state

Define the wording for accepted, submitted, settled, returned and unknown states in the actual product. Different teams may otherwise use completed to describe different events.

During a hypothetical provider timeout, the app might show failed while the external payment has settled. A customer may then submit another payment. The recovery design needs to prevent that avoidable duplication and correct the original status with appropriate evidence.

Test the customer-facing statement and downloadable records as well as the live screen.

Restoring one view while historical exports remain wrong can prolong the incident for finance teams using those exports.

Exercise the manual adjustment route

Some incidents require corrections. Specify who can initiate an adjustment, which evidence is required and who reviews it. Keep the original event and correction linked rather than rewriting history invisibly.

Use a realistic case volume. A process that needs fifteen minutes of senior review per payment becomes a major constraint across hundreds of affected transactions. Decide how cases are prioritised without skipping necessary controls.

After the exercise, compare the estimated recovery workload with available staff and service obligations. The outcome may justify better automated matching or clearer provider records rather than a faster application restart. Resilience investment becomes more precise when it addresses the work required to establish the correct financial position.

Finish with evidence of restoration

Close the incident only after agreed financial checks, unresolved-case ownership and customer communication are complete. Record the difference between technical availability and full service recovery.

Use the incident review to identify which work delayed dependable service: infrastructure recovery, transaction matching, corrections or customer communication. Fund that work. The next outage will test the complete payment operation, regardless of which dashboard turns green first.

A practical 90-day implementation plan

  1. Map the important business service from customer instruction through external settlement and balance display.
  2. Create deterministic states for accepted, rejected, unknown, duplicated and reversed work.
  3. Exercise timeout, partial dependency recovery, stale status and bulk manual adjustment scenarios.
  4. Close the exercise only after financial reconciliation, customer remediation and evidence review are complete.

Frequently asked questions

When is a payment service recovered?

When customers can complete the intended outcome and the firm can demonstrate correct financial state, settlement, communication and control - not merely when a technical endpoint responds.

Why freeze a financial position?

It prevents uncertain work from being replayed into a moving ledger and gives operations a reference point for classification and reconciliation.

What should impact reporting include?

Include affected customers, payment value, duration, vulnerable cohorts, delayed obligations, complaints, manual work and any residual remediation.

Reporting by Shoreditch Talk