Why most AI initiatives fail before they begin

The pattern is depressingly consistent. A senior team reads a case study, books a workshop, signs off on a tool, and three quarters later the dashboard is empty. The model is fine. The data is fine. The wrong question was asked.

1. What number, exactly, are you trying to move?

If the answer takes more than a sentence, you do not have a project — you have a hope. AI is a multiplier on a clear lever; without one, you are buying complexity, not capability.

2. Who eats it if it goes wrong?

An owner without consequences is a sponsor. Real owners pause Slack to read the eval results.

3. What is the manual baseline?

Before optimising with AI, time the human version. That number is the one your AI must beat — not a benchmark from a paper.

4. Where does this break?

The interesting failure modes are never in the demo. Run it on the messy 5%, the angry email, the legacy form. That is your real product.

5. Who tells you to stop?

Successful AI programmes have a clear off-ramp. If yours doesn't, you'll keep paying the meter long after the meter stopped meaning anything.

The companies winning with AI in 2026 are the ones who refused to confuse activity with movement.