AI Scrum Master
Illustrative scenario · Acme — 60-person product org
Acme adopts a draft-then-approve sprint plan
A scenario for a product org whose sprint planning has outgrown one Scrum Master: what to turn on, in what order, and which numbers tell you whether it worked.
- Team
- 60 people across 6 squads
- Cadence
- Two-week sprints, shared backlog per squad
- Existing tooling
- CSV export from the previous tracker
Acme is a fictitious team. This is not a customer story and contains no result figures or attributed quotes. What it does contain is real: every route below exists in the product, and every measure states where its number is computed.
Where it starts
- Planning takes a full day per squad and still ships over-committed sprints.
- One Scrum Master covers three squads, so two of them plan without one.
- Estimates are carried over from the last sprint whether or not they were right.
- Nobody can say which squads are over-loaded until the sprint already slipped.
What to turn on, in order
Order matters — each step assumes the one before it is already in place.
- 1
Import the backlog before anything else
The CSV importer maps columns onto the real task fields — status, priority, story points, assignee, sprint and due date. Planning quality is bounded by import quality, so this comes first and the importer reports rows it could not place.
- 2
Record capacity and skills per member
The draft plan reads open load, recent completed points, recorded meeting and PTO pressure, and the skills on each member's profile. Missing any of those produces a plan that looks confident and is not.
- 3
Turn on the draft plan for one squad
The AI Scrum Master produces a draft with a plain-language rationale per suggested task. Nothing reaches the board until a human accepts it, and accepting writes through the same task and sprint endpoints a manual edit uses — so the activity log and notifications look identical either way.
- 4
Set sprint goals the draft can be measured against
Goals give the health score something to compare against beyond raw throughput. Without them the gauge only tells you whether tasks moved.
- 5
Subscribe the right people to the right events
Notifications are per-user and per-type. Squad leads get draft-ready and slipped-item alerts; the rest of the org does not.
- 6
Run the retrospective against the actuals
Retros compare the committed plan with what shipped. That comparison is the feedback loop that makes the next draft better — skipping it leaves the plan learning nothing.
What to watch
No target values. Each row says where the number is computed and what feeds it, so you can decide what good looks like on your own data.
| Measure | Where the number comes from |
|---|---|
| Planning time per squad | Wall-clock time in the planning meeting, recorded by the squad. JiraVision does not time your meetings. |
| Committed versus completed points | Computed on /dashboard/sprints from the task rows in each sprint — the same numbers the burndown lines plot. |
| Draft acceptance rate | Accepted versus dismissed recommendations, on /dashboard/ai-scrum-master. Dismissals keep their reason, so a low rate is diagnosable rather than mysterious. |
| Load spread across the squad | The workload-distribution indicator on /dashboard/ethical-metrics, which prints its own formula and exclusions when you open the card. |
Walk this scenario on your own workspace
Every route above is live. The fit quiz will tell you which of these scenarios is closest to your team before you start.