How to run backlog refinement
Prepare and refine a PM-agreed subset of upcoming stories; preserve partial meeting coverage, per-story discussion and confirmation, and resume the same Activity without overwriting approved sources.
Product statusAvailable nowNotes and transcripts as files work today; recording through the meeting bot is beta.
Why it matters
The Activity remains the same conversation throughout. It can resume after the meeting, accept more transcript evidence later and return to any suggested topic without being treated as a new run.
Owner
Product Manager
Human + Agent
Participants · Agents
Trigger
A refinement session is due for the upcoming delivery horizon.
Required context
Optional context
Skills
Suggested conversation arc
- 1Propose eligible refinement candidatesCheck initiative relevance, delivery horizon and priority, current status, unresolved scope, existing preparation and source availability. Propose candidate stories with inclusion/exclusion reasons; never equate all open backlog items with meeting scope.Skills: Prepare task context · Prepare an agreed refinement scope
- 2Discuss and agree the refinement setReview proposed candidates with the PM, discuss priorities and missing context, and agree the ordered selected stories plus the meeting horizon. Allow removal, addition or deferral before preparation.
- 3Prepare the refinement documentFor selected stories, summarise current behaviour, desired outcome, relevant brain guidance, gaps and meeting decisions needed. Create a preparation document with an agenda and per-story references; do not pre-decide answers.Skills: Prepare an agreed refinement scope
- 4Hold the meeting and capture partial evidenceAccept meeting notes or transcript in one or more instalments. Map evidence to stable story references and record coverage as discussed, not discussed, ambiguous or awaiting more evidence. A partial transcript can unlock sufficiently evidenced stories without claiming the whole meeting is complete.
- 5Draft and verify each discussed storyUse the transcript and discussion to identify changes for all stories in this refinement. For every affected story, create a draft update against its current source baseline. Check acceptance criteria, consistency and relevant brain rules; leave uncovered or unclear stories pending and ask rather than inferring decisions from silence.Skills: Synthesise refinement decisions per story · Draft a Jira story · Review an artifact against agreed criteria
- 6Discuss and confirm each story separatelyFocus the Activity conversation on one story with its draft, diff, evidence and findings. Let the PM revise, confirm, reject or defer it independently. Confirm the exact final content and source target before applying; re-check a changed baseline and preserve every previously approved version. Approval of one story never approves the rest.
- 7Record recap and remaining follow-upsSummarise decisions by story, separately identify applied versus merely confirmed drafts and preserve unresolved or uncovered items with owner and next question. Request the PM’s close/continue decision when coverage is incomplete; this Activity does not create or manage tasks.Skills: Record decisions and follow-up
Output
Refinement scope and agenda before the session; per-story draft updates, coverage map and a decision recap after it.
Artifacts
Creates: Refinement scope
Updates: Jira story
What good looks like
- The selected scope is explicitly agreed or visibly pending.
- Every selected story appears in the coverage map.
- Partial evidence never creates implied approval or changes for absent stories.
- Applied updates and merely confirmed drafts are kept apart.
Quality gate
Destination
Usually next
Deep dives from the AI PM Lab
Articles that explain the thinking behind this Activity.
- AI product operating model: from managing work to managing context
- Structured AI workflows: your team doesn't need better prompts – it needs structure
- Claude Code for Product Managers: the Setup That Turns a Chat Window Into a Colleague
- The anatomy of a Jira ticket for AI product management in agentic engineering
Choose the path that matches your role.
Set the Product AI direction with us, or test the shared Product Brain on real work with Jira and code.