How to Migrate from PagerDuty Without Losing Alert Coverage

Start with the deadline and dependency map

Get the notice deadline, renewal date and access end date from your own agreement and account records. A community report about monthly or annual billing is not evidence of your terms. Give the operational migration enough time to fail a pilot and recover before that deadline.

Create one row per service, with source integrations, transformed fields, routing filters, schedule, overrides, notification channels, deduplication behavior, acknowledgement permissions and downstream automation. Include status-page and ticket workflows that could react to a test.

Map behavior, not just object names

Two products can use the word “schedule” while handling overlaps or overrides differently. PagerDuty's legacy schedule documentation (opens in a new tab) describes one model and distinguishes it from its newer shift-based experience. Identify which model your account uses before exporting or recreating anything.

Record unsupported features as gaps with owners. Do not replace a requirement with “we probably won't need that.”

Build a shadow path

Use a separate receiver and test responders for copied or synthetic events. Disable customer-facing automation on that test path. The existing system remains responsible for real pages while the replacement is evaluated.

Compare event identifiers, severity, service labels and timestamps. Exercise a duplicated trigger, a delayed recovery, a missing label and a deliberately unacknowledged incident. A successful webhook response proves ingestion only; use device delivery tests to prove the remainder.

Move one pilot service

Pick a service with a clear owner and reversible integration. Schedule the cutover, freeze unrelated routing changes, snapshot configuration and list currently open incidents. Define who owns those open incidents after the switch; do not assume their state transfers.

Write an ordered switch procedure for your integration: enable the new authoritative route, disable the previous production notifications at the appropriate boundary, and verify a test event. Rehearse the ordering in the pilot so it does not introduce an unobserved gap or duplicate response. There is no universally atomic switch across unrelated providers.

Rehearse rollback and expansion

Keep old credentials and configuration available for the agreed fallback period with appropriate access controls. Rollback restores the previous event destination and notification policy, then verifies one test event. Measure the time and record how you handle incidents created during the failed pilot.

Expand in small service groups only when the previous group has passed handoff, missed-acknowledgement and recovery tests and responders can operate the new interface. Save history and required records before closing the old account. Retire obsolete credentials after the fallback period ends.

This plan applies to renewal-driven exits and larger migrations without claiming that any replacement has feature parity with PagerDuty.

Compare four weeks of layered coverage

Export or record the effective schedule for the next four weeks from both systems. Compare the selected primary and backup at each handoff, at every temporary override boundary, and during known leave. Include a later daylight-saving transition when it falls outside that four-week sample. A match on today's on-call person is insufficient evidence for a recurring rotation.

Create a difference table with UTC instant, displayed local timezone, old selected person, new selected person and the reason for any intended change. For a bounded override, examine immediately before its start, at its start, inside it, immediately before its end and at its end. Check every service using that schedule rather than assuming one service test covers all routing.

PagerDuty's legacy model (opens in a new tab) gives later layers precedence. Its shift-based model (opens in a new tab) has different concepts for replacement and additional coverage. Identify the source model and map the resulting coverage, not only the object names or layer order.

Have the schedule owner approve intentional differences and block cutover on unexplained gaps. Keep the previous export and change record for rollback. A four-week comparison is a practical sample, not proof of every future holiday, membership change or timezone boundary; retain the schedule-change tests for later edits.