Translations:Affinity Estimating/7/en

Affinity Estimating with AI

When the backlog includes stories with a generative AI component, affinity estimating gets more complicated: two seemingly similar stories can turn out to be very different in size depending on data availability, the quality of the base model, or the acceptance criteria for the output. It's a good idea to create a separate grouping for AI stories until the team has its own velocity benchmarks for this kind of work.

Some collaborative digital whiteboard tools (Miro, FigJam) now include AI features that can suggest initial groupings. These can be useful as a starting point, but the manual Silent Sort has a value that goes beyond estimation: the process of sorting in silence activates individual thinking before the group discussion, something AI can't replicate.

Common mistake

Breaking the silence during the Silent Sort. If, during the individual sort, someone says out loud why they're placing a story in a particular spot, it contaminates everyone else's judgment. The value of this phase is precisely that each person forms their own opinion before being exposed to the group's.