The hardest part of an AI strategy is rarely a shortage of ideas. It is the opposite: a long list of plausible use cases and no defensible way to decide which one to fund first. Choose wrong and you spend a year proving that AI is hard. Choose well and you build the credibility to fund everything after it.

This article describes the scoring framework we use to turn a messy list of candidate use cases into a ranked, fundable order. It is deliberately simple — simple enough that a leadership team can apply it in a workshop and trust the result.

Three dimensions, scored honestly

We score every candidate use case on three dimensions. The discipline is not the scoring itself; it is being honest, especially about the dimension teams most like to skip.

Value

If this works, what changes, and is the change worth pursuing? Value can be cost saved, revenue enabled, risk reduced, or time returned to people — but it has to be expressible as something the business already measures. "Improves efficiency" is not a value score. "Removes roughly twelve hours of manual work per week from the claims team" is. If a use case cannot be tied to a real number, that is itself important information.

Feasibility

Can this actually be built, integrated, and operated with current technology and the team you have? A use case can be valuable and still be a poor first choice because it depends on systems that are hard to integrate, demands accuracy higher than current methods can reliably hit, or needs skills the organisation does not yet have.

Data readiness

This is the dimension teams most want to wave through, and the one that most often decides the outcome. Does the data this use case needs exist, is it accessible, and is it good enough? A brilliant, feasible use case sitting on data that is scattered, ungoverned, or poor quality is not a first project. It is a data project wearing an AI label.

A use case that cannot be tied to a number the business already measures is telling you something important.

Reading the scores

Score each candidate from one to five on all three dimensions. The pattern that emerges matters more than the arithmetic.

  • High on all three — your first project. Strong value, buildable, and the data is ready. This is what earns the right to do everything else.
  • High value, high feasibility, low data readiness — a real opportunity gated behind data work. Fund the data work first and revisit; do not start the AI build.
  • High value, low feasibility — keep on the roadmap, not the start line. Conditions may change; today it is a way to spend a year learning that AI is hard.
  • Low value — decline, regardless of how interesting it is. An elegant solution to a problem nobody is paying to solve does not build momentum.

Why the first one matters most

The framework is not really about finding the theoretically optimal use case. It is about choosing a first project that will succeed visibly — because the first project sets the organisation's belief about whether AI works here. A high-value, feasible, data-ready use case that ships and moves a real number does more for an AI strategy than three ambitious projects that stall.

Pick the use case that can win, win it in public, and use the credibility to fund the harder, more valuable work that needs it.