Make the model rank your backlog, not just describe it

A backlog of "everyone's pet project" usually survives a prioritisation workshop unchanged, because ranking things against each other out loud in a room means someone's project has to come last, and nobody wants to be the one who says so. Give a model the full list and a small set of explicit criteria, and ask for a strict rank order with no ties - not a grouping into "high/medium/low", which just recreates the same avoidance with three buckets instead of one list.

The useful part isn't the ranking itself, which you should treat as a draft to argue with, not a verdict. It's that a strict order with reasons attached gives the room something concrete to disagree with, which is far more productive than a vague debate about vibes. "Why is this eighth and not third" is a conversation that goes somewhere. "What should we prioritise" often doesn't.

A ranking with visible reasoning also survives a challenge better than a ranking with none. If the criteria and the reasoning per item are both shown, someone who disagrees can point at the specific input that's wrong ("customer impact for item 4 is actually higher than that") rather than rejecting the whole exercise. That's the difference between a prioritisation that gets refined and one that gets thrown out entirely because nobody can tell where it went wrong.