AI Use-Case Scorecard
Before you add an LLM to a feature, test the task. Answer six questions and get a verdict on fit, plus the highest autonomy level it is safe to start at.
What the six questions test
Each question scores 0, 1 or 2. The first two carry extra weight: if a wrong answer is costly or silent, or the output is hard to check, the task should be redesigned or kept with a person however good the rest looks. The autonomy level follows the lowest of the safety scores, because a system is only as autonomous as its weakest link.
The autonomy ladder runs from assist (the user does the work), to suggest (the model drafts and the user decides), to act with approval, to act within guardrails, to fully autonomous. Start low and climb only as acceptance and error data accumulate. The tool never recommends full autonomy as a starting point.
The full framework is in AI product strategy: deciding where LLMs belong in your product. For the design of safe behavior, read shipping AI features safely and designing agentic products users trust. Measure quality with evaluating LLM features, estimate the running cost with the LLM cost calculator, pick a model with the AI model selector, and place the decision inside your wider plan using how to write a product strategy.
Embed this tool
Free to use on your own site. Copy the code — the caption links back here.
Built by Deepanshu Grover, a Senior Product Strategy Consultant and AI builder in Paris. Runs entirely in your browser, nothing is sent anywhere. About · Get in touch