Public-Sector AI Use-Case Scorecard

Many public-sector AI projects stop after the demo. A model performs well in a workshop, but the project never becomes a system that a ministry trusts enough to run.

The model is usually not the main problem. The harder work is choosing which use cases are worth doing, assessing the risk honestly, and agreeing what "reliable enough to launch" means for that use case before anyone starts building.

I built a working tool that covers the first stage of that process: the work a product person does with the client before engineering starts. You scope a candidate use case (entity, problem, users, solution pattern), score it across five levers (mission impact, feasibility, data readiness, trust and risk exposure, time to value), and it computes a priority and plots it on a value-versus-feasibility quadrant.

The most useful part is what it produces next. It generates an evaluation plan (the specific metrics and the test data) that changes with the type of solution, because evaluating document extraction is very different from evaluating an agentic workflow or a forecasting model. And it produces a production-readiness gate that gets stricter as risk rises, up to human-in-the-loop review, red-teaming, and data residency for sovereign contexts.

I prototyped the whole thing hands-on in Claude Code. The scoring is simple and based on my own judgement, and it is not a validated model. Its purpose is to get the team to agree how reliable the system needs to be before they build it, which is a step these projects often skip.

A self-contained, working demonstration of how I would take a public-sector AI project from pilot to production. I built the tool myself rather than describing the approach in a document. You can try it in the live tool.