A roadmap workshop goes wrong when it starts with features. Start with the problems and the constraints, and let the features fall out of them — otherwise you spend the session defending pet projects.
Fill the board with problems first. Reacting and ranking is a second, shorter pass — mixing the two turns the workshop into a debate before the options are on the table.
Put team capacity, deadlines and known commitments on the board as cards too. A roadmap built without them is a wish list.
The AI summary captures the themes, the most supported items and the next steps, so the roadmap does not live only in the memory of whoever ran the session.
Keep it small enough to decide: product, engineering and one person who talks to customers daily. Anyone else can add cards to the board asynchronously before the session.
Use reactions to see what the group supports, then rank the top items against effort and the constraints on the board. The summary records the reasoning so the decision holds up a month later.