The prioritization matrix
| Dimension | Scale | Question |
|---|---|---|
| Value | 1–5 | Frequency × effort × outcome value |
| AI fit | 1–5 | Suitability for probabilistic interpretation/generation |
| Readiness | 1–5 | Data, systems and owner availability |
| Risk | 1–5 | Consequence of incorrect output |
| Effort | 1–5 | Integration, change and maintenance |
| Confidence | Low / Med / High | How reliable the estimates are |
Do not over-engineer the weighting
Use the scores to structure a decision, not simulate certainty. A workshop conversation about why two people scored risk differently is often more valuable than a complicated formula.
Add dependencies and sequence
A use case can be attractive but blocked by another capability. Add dependency notes so the final output is a roadmap rather than a static rank.
Frequently asked questions
Should I calculate one total AI score?
You can, but keep the component scores visible. A single number can hide unacceptable risk or weak evidence.
What is a good first AI use case?
One with meaningful value, strong fit/readiness, low or controllable consequence of error, and a clear owner.
Should strategic importance be a separate dimension?
Yes if it materially changes investment priorities. Just distinguish strategic value from measured operational value.