An integration can be online while quietly dropping records, duplicating updates, or delivering stale data. Monitoring should show whether the business workflow is healthy, not only whether an endpoint answered.
Define the business event
Record what starts the integration, what should arrive, how quickly, in which order, and what a successful result looks like. Include normal, delayed, missing, duplicate, and partial states.
Monitor delivery and health
Track requests, successes, failures, latency, retries, queue depth, authentication errors, rate limits, provider status, and the last successful business event. Use baselines rather than arbitrary thresholds.
Check data quality
Compare counts, required fields, identifiers, statuses, timestamps, relationships, and source-of-truth rules. A successful API response does not prove that the destination record is usable.
The integration discovery checklist and uptime SLA checklist help define ownership.
Make alerts actionable
Send alerts to the person who can investigate, include event, system, record, time, error, retry state, and next action, and avoid exposing unnecessary sensitive data. Escalate unresolved failures.
Test recovery
Practice retry, replay, deduplication, credential rotation, provider outage, partial success, manual correction, and rollback. Record the incident and improve the runbook.
Integrations failing silently between reports? Ask Vertinus to define the health and exception checks.