AI Use Case Scorecard
Most AI output problems start before you open the tool.
The AI Use Case Scorecard helps you check whether a task is a good AI candidate, in 10 minutes, using whatever tools you've been given.
If you've been told to adopt AI tools and the output keeps disappointing you, the instinct is to blame the prompt. Rewrite it. Try a different phrasing. Ask a colleague how they get better results.
Most AI output problems start at use case selection. The task was a poor candidate for AI from the beginning: it required judgment the tool couldn't apply, context it couldn't hold, or verifiability that doesn't exist. No prompt fixes that. The output was always going to need heavy correction.
The scorecard gives you a way to check before you start. Six criteria, a score out of 18, and a clear reading on whether the conditions are right to run an experiment or whether you need to reconsider the task first.
It takes 10 minutes per use case. You can use it on any task, with any tool, today.
What's in the scorecard
Six criteria that determine whether a task is a good AI candidate:
Risk level: how bad is it if the output is wrong?
Reversibility: how easy is it to catch and fix a mistake?
Context requirements: how much must you know to verify what the tool produces?
Judgment load: does this task require decisions the tool can't make?
Volume and repetition: does scale actually help here?
Verifiability: can you tell quickly whether the output is correct?
Score each criterion from 1 to 3. Your total score tells you whether you're looking at a strong candidate, a marginal one worth a small experiment, or a poor fit where the correction cost will exceed the time saved.
The scorecard explains what to do with a low score. When a task scores low, the use case is the problem, and the scorecard tells you where to look next.
Who this is for
This scorecard is for you if:
You've been mandated to adopt AI tools and you're not confident you're using them on the right tasks
You're spending more time correcting AI output than it saves, and you don't know if that's a prompting problem or a use case problem
You want something more rigorous than impressions to evaluate whether your tools are working
You want a starting point that doesn't require overhauling your whole testing process
If you've ever thought "this is producing slop regardless of how I phrase it", this scorecard will tell you whether the task was ever a good fit.
The scorecard is one page, it takes 10 minutes, and you could use it on a real task today.
I'll send it to you by email. If you want to hear when the full evaluation framework opens for enrolment in May, you'll be the first to know.