When a company says it wants to “use AI,” the usual starting question is “what can AI do for us?” It is too broad and produces aimless pilots. The more useful question is: which role, making which decision, would change because of data or a model — and would that change be measurable?
Three screens for where value lives
- Measurability. If you cannot quantify the improvement, it is not the right first target.
- Decision frequency. A decision made daily is worth automating more than one made yearly.
- Data readiness. Is the needed data already present, clean, and permitted for use?
Where it usually does not pay
Projects that sound exciting but never return share a pattern: the decision-maker does not trust the model, the improvement cannot be attributed, or the real change is process, not algorithm. Naming those and deferring them is itself a deliverable.
We recommend starting with a diagnosis that runs candidate scenarios through these three screens — and piloting only the one that passes.