AI adoption in the first 100 days: what an investor should ask.

An AI application can work on day one and still have delivered nothing by day one hundred. The technology does what it should, while the people it was built for keep working the way they did. For an investor, that gap matters more than the model: the value case assumes the work changes.

McKinsey’s rule of thumb is that every amount spent building a digital solution needs at least the same again for process change, training and change management. These five questions test whether that second half is in the plan.

1. Who has been freed up to work on it?

Engineering capacity can be brought in. Knowledge of how the work is actually done cannot. An initiative without time from the people who do the work becomes a side project, and side projects meet resistance when they arrive as the new way of working. Ask how many hours a week the business has committed, and who covers their daily work in the meantime.

2. Is the application built beside the process, or on top of it?

Adding AI to a process that was designed without it often changes little. A small team that rebuilds one step from scratch, next to the existing process, can move faster. The harder part comes later: bringing what that team learned back into the organisation. Ask how that return is planned, before the pilot starts.

3. How is use measured, every week?

Licence counts and satisfaction scores say little. Three measures say more: how many of the intended users work with it each week, what share of the process runs through it, and who stops and why. Ask to see them weekly, not at the end.

4. What decides whether the next team follows?

Scaling should follow evidence, not the calendar. Agree a threshold for use in the first two weeks. When use reaches it, the next team follows. When it does not, the reasons are the most valuable output of the pilot.

5. What will the first experience be like?

Teams tend to give a new tool one real chance. A first version that fails in daily work is hard to reintroduce, however good the next one is. Ask when the application will reach its users, and what has to be true before it does.

At day 100.

A useful review covers results against the baseline, use per team and what did not work. The last part is the one most often left out, and the one an investor learns most from.