Two working APIs can still produce a failing business process. The gap usually appears between technical success and operational completion.
A request returns 200, but the downstream system later rejects the record. A retry creates a duplicate. A status means different things in each application. Nobody owns reconciliation. Each system is healthy while the workflow is broken.
What good engineering looks like
Design integration around business completion. Use correlation identifiers, idempotency, explicit state mapping and measurable service levels. Assign owners for both the interface and the cross-system outcome.
-
Shared definition of completed work.
-
State and error mapping across systems.
-
Idempotent retries and duplicate protection.
-
Reconciliation for delayed or missing records.
-
Operational dashboards and escalation ownership.
A practical starting point
-
Trace one failed transaction across all systems.
-
Separate transport, data, decision and ownership failures.
-
Add one end-to-end completion measure.
-
Run a recovery exercise with the responsible teams.
The decision to make
Integration is dependable when the organisation can detect, explain and recover the business outcome, not only when endpoints respond.