How to know if your AI project is working

Most AI projects fail quietly. Nothing crashes — the tool just stops being used, and six months later nobody remembers why it was built. The difference between projects that stick and projects that fade is almost always the same thing: a number, chosen before anything was built.

One number, chosen early

Before we write a line of code, we agree with the client on the single number the project exists to move. For a clinic we worked with, it was missed appointments. For a logistics operator, on-time deliveries. Not a dashboard of fifteen KPIs — one number that the owner already cares about.

Measure before, not just after

You cannot know a tool saved 45 minutes a day if nobody measured the morning routine before it existed. The baseline is half the measurement, and it is the half everyone skips. Week one of our projects is partly spent writing down how long things take today.

Adoption is the leading indicator

The business number moves slowly. Usage moves fast. If the team stops opening the tool in week two, the business number will tell you the same thing in month three — too late. We watch adoption weekly and treat a drop as a bug, because it is one: either the tool does not fit the workflow, or the workflow was misunderstood.

A simple rule to steal: no AI project without a number, a baseline, and a person who checks both every month. It takes one meeting to set up, and it is the cheapest insurance a project can have.