A 200 response is the easy part

The difficult work starts after the demonstration. Production integrations must handle timeouts, partial failures and repeated requests without creating incorrect business outcomes.

Reliability comes from designing for failure rather than treating failure as an edge case.

Authentication, validation and ownership determine reliability

Tokens expire, credentials change and external services reject data. Each failure should produce a controlled response and a useful record.

Every integration also needs a named owner who receives alerts and understands the recovery process.

Retries must be safe

A timeout does not always mean the external system failed. It may have processed the request before the response was lost.

Idempotency prevents the same business transaction from creating duplicate orders, shipments, invoices or payments when the request is repeated.

Silent failures become operational debt

Structured logging should record the request reference, status, timing and error without exposing sensitive data.

Monitoring should turn repeated failures, expired credentials and unusual response times into visible alerts.

Integration maintenance is ongoing

Third-party APIs change versions, fields, authentication methods and rate limits.

A reliable integration needs monitoring, documentation, ownership and an agreed maintenance path after launch.

Business example: duplicate courier shipments

Before

A booking request times out, the user retries and the courier platform creates a second shipment because the first request had already completed.

→
After

The integration sends an idempotency key, checks the original transaction reference and returns the existing shipment instead of creating another.

A practical comparison

Higher-risk approach

  • Single test call
  • Blind retries
  • Minimal logs
  • Manual token replacement
  • No named owner

Controlled approach

  • Failure-path testing
  • Idempotent retries
  • Structured monitoring
  • Credential lifecycle
  • Defined ownership
A successful test call proves the integration can work. Reliability proves it can keep working when conditions are not ideal.