Start with the queue, not the model

Every project we turn down starts the same way: a team picked a model, then went looking for work to give it. The projects that pay back start from the other end.

Look for the resented queue

Ask any team which inbox, board or spreadsheet they dread on Monday. If it fills every day, follows rules most of the time and has a clear definition of done, you have a candidate.

Count before you build

We spend the first week counting: volume per day, minutes per item, and the share that follows the rules. If fewer than half the items are routine, an agent won't pay back yet.

Then pick the model

Once the queue is measured, the model choice is usually boring. That's a good sign.