A consultancy's return criteria rank agent projects by the size of the model each one would call, so the largest model wins the funding. The process each project would improve is not in the score. What should the criteria score instead?
Select an answer to reveal the explanation.
Short Explanation
Hiring the biggest truck is not the same as moving the most freight. You score the process benefit the project produces, and you stop treating model size as the value. A larger model can still automate a process the client does not need.
Full Explanation
Return criteria exist to compare what the business gains, which is the process benefit, with what the work costs. Model size is a design choice inside a project, not the benefit the consultancy should fund. Scoring the process benefit lets a small, well-aimed project beat a large model pointed at a weak process. Ranking by parameter count keeps the current mistake and will fund the heaviest model even when the client process barely moves. Counting topics measures how much conversation was built, not whether a process got faster or more accurate, so a long topic list can outrank a thin agent that removes a real bottleneck. Scoring prompt length treats authoring volume as value, and a long instruction can still describe a process that does not pay off. Exam caveat: the model can still appear on the cost side of the case. It does not belong on the benefit side as a substitute for the process result. Operational check: rewrite the ranking sheet so each project has a named process benefit, and confirm a project cannot win on model size while that benefit line is blank.