The first AI pilot should prove savings, not ambition

A lot of AI transformation work starts too big. The company wants a platform. The team wants a roadmap. The consultants want a full operating model. Everyone wants to prove they are thinking strategically.

That is understandable. It is also usually the wrong first move. The first AI pilot should not be a vision exercise. It should prove savings.

Pick one workflow that is manual, repetitive, measurable, and painful enough that people already know it is broken. Then build the smallest version of the automation that can do real work under human approval. Not a demo. Not a deck. Not a chatbot that answers questions around the work. The actual work.

Most companies do not fail at AI because they lack imagination. They fail because they start with ambition before they have proof.

Local first, modular second, platform later

Local first means the pilot is close to the operator who feels the pain. It uses the language, constraints, approvals, exceptions, and handoffs of the actual business.

Modular second means the automation is built as a reusable operating unit. If it works in one place, you can adapt it to the next workflow, market, or team without starting from scratch.

Platform later means you earn the right to standardize after the business has seen measurable relief.

AI does not remove complexity. It exposes it. That is why the best pilots force the hard questions early:

• What exact task is being done?
• Who approves the output?
• What happens when the case is unusual?
• How do we know the automation saved time or reduced cost?
• What must stay under human control?
• If this works, what is the next workflow we would seal?

Why "sealed" matters

We do not trust wild automation in real operations. We trust a workflow where AI proposes, humans approve, and the approved process runs the same way every time until someone intentionally changes it.

That is how you get accountability. It is also how you get economics: when the workflow is sealed, you can measure before and after. Hours removed. Backlog reduced. Turnaround improved. Errors lowered.

Once you have that proof, the broader conversation gets easier. You are no longer asking people to believe in AI. You are showing them a working piece of the future inside their own business.

The first pilot should be small enough to trust, specific enough to measure, and valuable enough that the business wants the next one.

Also shared on LinkedIn · read the original