Here’s a scoring sheet for picking which AI use cases to pilot. You score each candidate 1 to 5 on eight criteria, and the workbook weights them and puts each one in a band: pilot it, fix the gap first, or not yet. Half the weight sits on the four things a pilot never tests: whether you can handle the exceptions, whether a named team will own it, whether there’s a tested way back if it fails, and whether you can defend its decisions. Raw technical feasibility is only 15%.
Keep it to ten candidates or fewer, set aside an hour, and score what’s true today. “We’re hiring for that” is a 2. Bring the person who’d own each one in production, because without them the owner and off-ramp scores are guesses.
The How to use tab has the anchors for every score, and the Sample tab has six candidates scored by an insurer’s operations team. The customer email agent that demos best lands mid-pack, which is the point of the exercise. Prefer pen and paper? The PDF has the same anchors, a blank grid, and the worked example.