Challenge

Customer onboarding, refunds, and provisioning still required manual hops across spreadsheets, email, and five SaaS tools. Engineers carried pagers for brittle nightly scripts; business users could not see why a case stalled for days.

Context

Each team had solved a slice with Zapier-like tools or one-off Lambda functions, but there was no shared pattern for retries, idempotency keys, or compensating transactions when a downstream API changed.

Approach

• Introduced a canonical event bus with versioned schemas, dead-letter queues, and consumer lag dashboards.
• Built a connector kit for Salesforce, Stripe, Zendesk, and internal REST services with shared backoff, signing, and secret rotation policies.
• Shipped an operator console for replay from arbitrary steps, safe pause during incidents, and exportable run history for auditors.
• Added synthetic monitors that drive end-to-end happy-path and failure-path scenarios every hour.

Reliability

We documented SLOs per workflow, error budgets, and escalation paths. Runbooks linked directly from alert payloads so on-call could act without tribal knowledge.

Outcome

Shipping a new cross-system automation dropped from multi-week projects to days. Silent failures largely disappeared; operations could explain status to customers with trace-backed evidence.