PI Planning is the ART's planning meeting with the cameras on. Two days, every team in the same (physical or virtual) room, a draft plan that gets punched in public, and a confidence vote that is worthless if people are punished for voting two. Done well, it is cheaper than twelve weeks of quiet collision.
Do the unglamorous work before day one
PI Planning fails in the prep, not in the icebreaker. Product Management needs a ranked program backlog of features that could actually fit a PI — sized, split, and with acceptance that is not 'the steering group likes it.' Architecture needs a short briefing on what must be true (and what is still a bet). The RTE needs a room or a tool setup that lets teams see one program board, not seven personal Miro boards that will never be merged.
Invite the people who can change the plan: Product Owners, Scrum Masters or team coaches, tech leads, business owners who will later accept PI objectives, and the architects who will otherwise appear in week six as a surprise constraint. A planning event of only delivery staff produces a schedule that leadership will overwrite on Monday. That is not alignment. That is a dress rehearsal for the real decisions, held without the decision-makers.
Send a pre-read and a capacity reality check. Teams should know holidays, support load, and reserved work before they forecast. A team that 'commits' at 100% of theoretical hours is writing fiction with better markers. If capacity is unknown, stop and get it. Planning without capacity is how you leave with a beautiful board and a PI that was already late on the way out of the car park.
What the two days are for
The opening is context, not theatre: why this PI matters, what the product vision is, what the architecture will and will not support. Then teams break out. They pull features, draft sprint-level plans, write PI objectives in language a business owner can accept, and put dependencies on a shared program board. The board is the product of the event. If dependencies live only in someone's notes, you held a workshop, not a planning PI.
Draft-plan review is where honesty earns its keep. Teams present what they think they can do and what they need from others. Management and business owners do not 'motivate' the room by adding scope. They resolve conflicts: this feature over that one, this dependency moved, this objective uncommitted because the vendor date is a wish. A review that only applauds is how you discover in week three that three teams planned the same integration week and none of them planned the integration.
Day two adjusts, then locks a version of the truth. Risks get ROAMed — Resolved, Owned, Accepted, or Mitigated — with names, not vibes. Teams vote confidence in the PI plan. A low vote is a finding. If the room always votes five because the last person who voted two was taken aside, you have a climate problem, not a plan. End with PI objectives that a business owner will actually sign, including uncommitted objectives for the work that is a bet.
After: the plan is a hypothesis with owners
Publish the program board, the PI objectives, and the ROAM list within a day. Owners confirm dependencies they accepted. A dependency that nobody confirmed is not a dependency. It is a hope that will become an issue with better lighting. Book the first ART sync and the mid-PI inspection now, while calendars are still ashamed of themselves.
Protect the Innovation and Planning iteration if you have one. It is for hardening, innovation, and the next planning prep — not a secret sprint you fill because someone promised a date in the corridor. Burning the IP iteration is how trains lose the only slack that made the next PI planning sane.
When the first week proves a feature will not fit, change the plan in the open. PI Planning is not a contract with the universe. It is a baseline you inspect. Quietly dropping an objective so the confidence vote still 'looks good' is how the next planning event starts from a lie.
Summary & next steps
Prep the backlog, capacity, and the people who can decide. Use the two days to produce a program board, PI objectives, and ROAMed risks. Treat a low confidence vote as data. Publish owners within a day.
Next: cross-team dependencies — how to see them, shrink them, and stop pretending a red string is a delivery strategy.