Your first 90 days with an enterprise AI mandate

The appointment came with a title, a board expectation, and remarkably little else. There is no map, the previous owner of the initiative has moved on or never existed, and the first question at the next executive meeting will be some version of "where are we with AI."

The role needs two things at once, and they pull in different directions: an early win that is defensible, and external validation that the direction is right. The sequence below is built to produce both.

Days 1–30 · Find the number

Do not commit to a roadmap this month. Spend it finding the workflows where AI would move a number somebody already reports on.

• Interview the P&L owners, not the technologists. Ask what their team spends time on that nobody would defend in a board meeting.
• Audit what already exists — shadow tools, abandoned pilots, licences nobody uses. There is usually more than anyone admits.
• Refuse, politely and repeatedly, to commit to an enterprise-wide strategy this month. The pressure to do so is the first trap.

Days 31–60 · Pick one workflow

The instinct is to pick the most impressive candidate. Pick the most defensible one instead — the criteria differ.

• A named owner whose number moves. Without this, success is unattributable and therefore unfundable.
• A baseline you can measure this week, before anything changes.
• Operators who are willing. A hostile team can stall a working tool indefinitely and be entirely justified in doing so.
• A governance tier you can clear in weeks, not quarters. Save the hardest risk surface for the second win.

Days 61–90 · Produce visible proof

Boards and departments count different things as proof. The department wants to see the workflow change. The board wants a number, a cost, and a repeatable path. Produce both, in that order, and do not present the second without the first.

An early win that nobody outside your team can verify is not a win. It is a claim.

The traps

• The enterprise-wide strategy deck. Consumes a quarter, ages badly, changes nothing.
• The tool census. Feels like progress, produces a spreadsheet, ends in a licensing negotiation rather than a deployment.
• The centre of excellence with no delivery mandate. Becomes a review board that slows everyone else down and ships nothing itself.

What to have on a slide at day 90

One workflow, named. The baseline before, and the number now. Who runs it, and whether the vendor is still in the room. What it cost. And the next two workflows in the queue, with the same structure and no adjectives.