01 the business
Ecommerce platform
60 staff, 8 engineers
When something breaks, the engineers fixing it are also the ones asked for updates. Support hears nothing, the status page goes stale, and the write-up afterwards is rebuilt from memory.
02 before
The problem
When something breaks, the people who can fix it are also the people being asked for updates. Support hears nothing, customers see a stale status page, and the write-up afterwards is rebuilt from memory. Small teams rarely have a dedicated incident lead to handle the talking.
- Updates are irregular and late.
- Support and account managers are left guessing.
- Post-incident reviews take days and miss details.
03 the build
What the AI does
- Step 1: Opens an incident record when PagerDuty or Opsgenie fires and starts a dedicated Slack or Teams channel.
- Step 2: Reads the channel, linked Jira tickets and Datadog alerts to keep a running timeline of what is known.
- Step 3: Drafts a plain-English update for staff and a shorter one for customers at a set interval.
- Step 4: Posts approved customer updates to Statuspage and sends the internal version to support and account managers.
- Step 5: Writes a first draft of the post-incident review from the timeline once the incident is closed.
paused: waiting for a person to approve
How it was put together
- Opened an incident record and a dedicated channel when an alert fired.
- Kept a running timeline from the channel, tickets and monitoring alerts.
- Drafted updates for staff and customers at a set interval.
- Wrote a first draft of the post-incident review from the timeline.
04 human in the loop
Where a person stays in charge
The incident lead approves every customer-facing update with one click before it goes to Statuspage. The post-incident review stays a draft until the engineering owner edits and publishes it.
- The incident lead approves every customer update with one click.
- The review stays a draft until the engineering owner publishes it.
- Updates state only what is confirmed.
05 rollout
From kickoff to live in 3 weeks
- Week 1Connect alerting, chat and the status page; agree update templates.
- Week 2Run a practice incident.
- Week 3Use for all incidents above a set severity.
06 after
What changes
Engineers keep their heads down on the fix while customers and support stay informed. Updates arrive on schedule in consistent language, and the review starts from an accurate timeline instead of a blank page.
Assume two significant incidents a month, each needing 1.5 hours of an engineer's time on updates and 4 hours on the write-up.
2 × (1.5 + 4) hours = 11 hours a month of engineering time, mostly during the most stressful moments.
What to measure
- Time to first customer update.
- Update interval kept during incidents.
- Time to publish the post-incident review.
- Support tickets asking for status during incidents.
07 evidence
What the research says
The global average cost of a data breach reached $4.99 million, a record high.
IBM, Cost of a Data Breach Report (opens in a new tab) · 2026
In a study of 453 professionals doing writing tasks from their own jobs, access to ChatGPT cut the time taken by 40% and raised quality, as rated by evaluators, by 18%.
08 lessons
Mistakes to avoid
- Speculating in updates. Say what is known.
- Auto-posting to customers. Keep the approval.
- Skipping the review. That is where you prevent the next one.
09 questions
Common questions
- Which tools does it connect?
- PagerDuty or Opsgenie, Slack or Teams, Jira, Datadog and Statuspage.
- Will it diagnose the incident?
- No. It records and communicates. Engineers diagnose and fix.