It support · Production
incident.io builds AI-powered incident summarization: lessons from the first AI product
The problem
During active incidents, responders focused on mitigation found it painful to keep the incident summary current, leaving summaries stale and new responders without an up-to-date picture.
Workflow diagram · grounded in source
1
User provides incident update
trigger
“whenever you provide an update (which is a message about "what's happening" that gets broadcasted to the incident Slack channel), we'll use OpenAI's models to suggest a new summary of the incident”
2
OpenAI suggests new summary
ai_action
“we'll use OpenAI's models to suggest a new summary of the incident. This suggestion will consider your update, the conversation in the channel, and the previous summary (if there was one)”
3
User accepts or edits suggestion
human_review
“You can choose to accept it as-is or apply some tweaks first”
4
Feedback buttons capture signal
feedback_loop
“The first two buttons let the user accept/edit the suggestion, but the third button of Bad sugggestion is purely so we can track how often we were unhelpful. Not only did this mean we could track % acceptance / rejection and see ourselve…”
Reported outcome
Half of all summary updates are now written by AI, the launch reaction was better than the team could have hoped for, and there have been lots of requests for similar functionality across the product.
Reported metrics
summary updates written by AIHalf of all summary updates are now written by AI
Time saved for incident respondersloads of time saved
Reported stack
OpenAISlack
Frequently asked questions
What did this team achieve with this AI workflow?
Half of all summary updates are now written by AI, the launch reaction was better than the team could have hoped for, and there have been lots of requests for similar functionality across the product.
What tools did this team use?
OpenAI, Slack.
What results were reported?
summary updates written by AI: Half of all summary updates are now written by AI; Time saved for incident responders: loads of time saved (source-reported, not independently verified).
How is this it support AI workflow structured?
User provides incident update → OpenAI suggests new summary → User accepts or edits suggestion → Feedback buttons capture signal.