Most organisations don't have a shortage of AI ideas. They have a shortage of ways to tell which ones are worth doing first. This guide lays out a simple way to score and sequence use cases so the first projects you fund are the ones most likely to matter.

Start with the work, not with the AI

Before scoring anything, list where time actually goes: the repetitive tasks, the manual handoffs, the decisions that wait on someone digging through documents or systems. Use-case lists built from technology capabilities ("we could summarise things", "we could add a chatbot") tend to produce pilots. Lists built from where the business loses time or money tend to produce results.

The best AI roadmap starts as a work list, not a technology list.

Score every idea on two axes

  • Business value: time saved, revenue protected or created, risk reduced, or a bottleneck removed elsewhere in the business.
  • Feasibility: how ready the data is, how well-defined the process is, and how much human oversight the task realistically needs.

Plot the four quadrants

  • Quick wins — high value, high feasibility. Fund these first; they build the case for everything after.
  • Strategic bets — high value, low feasibility. Worth investing in fixing the blocker (usually data or process), not worth abandoning.
  • Nice to have — low value, high feasibility. Easy, but shouldn't be first in line for budget or attention.
  • Not yet — low value, low feasibility. Revisit later; don't spend this quarter's capacity here.

Turn the list into a roadmap

Once ideas are scored, sequence rather than prioritise everything at once. Start with one or two quick wins to prove the operating model — ownership, data access, measurement — actually works in your organisation. Use what you learn there to make the strategic bets more feasible, rather than tackling them cold.

A short exercise you can run this week

Gather five to ten people from the team closest to the work. Spend thirty minutes listing where time or accuracy is being lost today, without mentioning AI at all. Only then score each item against the two axes above. The exercise usually surfaces two or three quick wins that were never on anyone's technology roadmap, because they didn't start as a technology conversation.