Skip to content
← Tools
Integration & APIs

Migration risk register

Twenty risks that materialise on integration migrations, each with an early warning sign. Score them for your estate and export.

Twenty risks that materialise on integration migrations often enough to be worth pre-writing. Default likelihood and impact are ours; override both for your estate, exclude what does not apply, and export the result.

In register
20
High
8
Medium
12
Low
0
Excluded
0
  • ScopeThe interface inventory is incompletehigh · 20

    Interfaces surface after the plan is baselined, usually from a team that was not in the workshops.

    Early warning
    Someone says 'there's also the one that…' in a status meeting.

    Mitigation
    Reconcile the inventory against platform logs and firewall rules, not against memory. Budget contingency as a percentage of interface count, not of effort.

  • Third partyPartner re-certification takes longer than plannedhigh · 20

    Trading partners must retest and approve on their own change calendar, which is not yours to compress.

    Early warning
    The first partner takes more than two weeks to respond to the initial contact.

    Mitigation
    Open the conversation in week one regardless of when the build lands. Sequence partner work by their responsiveness, not by your convenience.

  • DeliveryTest data is unavailable or unrepresentativehigh · 16

    Testing proves the happy path because that is the only data anyone can produce.

    Early warning
    Test cases all use the same three records.

    Mitigation
    Capture and mask a slice of production traffic early. Include the awkward records deliberately — long names, negative quantities, missing optional fields.

  • DeliveryNon-production does not resemble productionhigh · 16

    Integration points are stubbed, volumes are tiny, and configuration differs.

    Early warning
    A defect is described as 'works in test'.

    Mitigation
    Treat environment parity as a deliverable of the foundation phase with its own acceptance criteria.

  • TechnicalCustom code has no equivalent on the targethigh · 15

    Java services, scripts or a transformation language the new platform cannot express.

    Early warning
    A capability comparison produces the phrase 'we can work around that'.

    Mitigation
    Identify these before platform selection, not after. They often justify keeping a capability where it is rather than moving it.

  • OperationalRollback has never been testedhigh · 15

    A documented rollback that has not been rehearsed is an assumption, not a plan.

    Early warning
    The rollback step in the runbook has no duration against it.

    Mitigation
    Rehearse rollback in a non-production cutover. Time it. If it exceeds the window, it is not a rollback plan.

  • DeliveryLegacy platform expertise leaveshigh · 15

    The one person who understands the old estate resigns or is reassigned.

    Early warning
    Only one name appears against the legacy platform in every conversation.

    Mitigation
    Extract knowledge into the inventory and runbooks early, while there is no urgency. Pair them with someone from day one.

  • CommercialThe old platform is never switched offhigh · 15

    A handful of interfaces are too hard to move, so both platforms run permanently and the saving never lands.

    Early warning
    The hardest interfaces keep moving to a later wave.

    Mitigation
    Do the hardest interfaces early, while contingency exists. Name a person accountable for decommissioning with a date.

  • Third partyPartner allow-lists still point at the old platformmedium · 12

    Egress addresses change with the platform and partners must update firewall rules.

    Early warning
    Nobody can produce the list of partners who allow-list you.

    Mitigation
    Collect egress requirements before the pilot. Where possible keep a stable outbound address so partners never have to change.

  • TechnicalConfiguration is hard-coded inside interfacesmedium · 12

    Endpoints, credentials and business rules set inline rather than externalised, discovered on promotion.

    Early warning
    The same interface behaves differently in two environments and nobody can say why.

    Mitigation
    Scan for literals before estimating. Externalising configuration is migration work that is usually left out of the estimate entirely.

  • TechnicalSubtle behaviour differences reach productionmedium · 12

    Rounding, null handling, date formats, character encoding and ordering differ between platforms.

    Early warning
    Output is compared by eye rather than by diff.

    Mitigation
    Run old and new in parallel against the same input and compare programmatically. Manual comparison finds the obvious differences only.

  • TechnicalBehaviour differs at production volumemedium · 12

    Timeouts, memory limits, throttling and batch sizes that never appear in testing.

    Early warning
    Peak volumes are unknown or quoted as averages.

    Mitigation
    Load test the largest interfaces at peak plus headroom, before cutover rather than after.

  • OperationalA certificate or key expires mid-migrationmedium · 12

    Expiry lands during the programme and is initially diagnosed as a migration defect.

    Early warning
    There is no inventory of certificates with expiry dates and owners.

    Mitigation
    Build the certificate inventory during discovery and renew anything expiring inside the programme window plus three months.

  • OperationalThe cutover window is too shortmedium · 12

    The plan assumes everything goes right, and the go/no-go point arrives after the point of no return.

    Early warning
    The runbook has no explicit go/no-go time.

    Mitigation
    Set the go/no-go decision at a point where rollback still fits in the remaining window, and hold to it.

  • OperationalMonitoring is not migrated with the interfacesmedium · 12

    Interfaces move but alerting, dashboards and log routing stay pointed at the old platform.

    Early warning
    The monitoring workstream has no owner.

    Mitigation
    Treat monitoring as part of the definition of done for each interface, not as a separate phase.

  • ScopeUndocumented downstream consumersmedium · 12

    Someone reads a file, a table or a queue that was never considered an interface.

    Early warning
    A staging table or shared folder has more readers than the design shows.

    Mitigation
    Trace consumers from the data, not the documentation. Announce decommissioning early and loudly enough that they surface.

  • DeliverySource system teams are unavailable when neededmedium · 12

    Migration needs their time for testing and cutover, which is not in their plan.

    Early warning
    Their commitment exists in your plan but not in theirs.

    Mitigation
    Get named people and dates into their planning cycle, not yours. Sequence waves around their availability.

  • CommercialDual-running costs are not budgetedmedium · 12

    Both platforms are licensed and supported in parallel, sometimes for a year.

    Early warning
    The business case shows the old platform cost ending at go-live.

    Mitigation
    Budget the overlap explicitly and negotiate licence terms for it before signing the target platform.

  • CommercialA wave collides with a business freezemedium · 9

    Year end, peak season or a regulatory window blocks change for weeks.

    Early warning
    The plan has no freeze periods marked on it.

    Mitigation
    Map freezes before sequencing. A collision costs a quarter, not a sprint.

  • ScopeDead interfaces are migrated anywaymedium · 8

    Nobody will confirm an interface is unused, so it gets migrated to be safe.

    Early warning
    An interface has no owner and no message in the last quarter, but is still in scope.

    Mitigation
    Turn it off in place for a fixed period before the migration and see who complains. Migrating it is the more expensive way to find out.

A generic register is not a risk assessment. Twenty risks that could happen is a starting list; knowing which three will actually bite your estate needs someone who has seen it, and usually needs the inventory first. Talk to us about yours.

Every risk here carries an early warning sign, because a risk you can only detect after it has happened is not a risk — it is an issue with better paperwork. Nothing is stored; export as you go.