An AI plan is easy to write. Knowing what it’s for is the work.

The moment

A board paper is due, and the question has a date on it now.

It has usually been circling for a year. Somebody read something, somebody else saw a competitor’s press release, and the item sat politely under any other business for three or four meetings. Then it moved up the agenda, and now there is a slot for it, and the slot has your name against it.

The awkwardness is not that you have no answer. It is that you have a great many possible answers, all of them defensible, none of them obviously more important than the others. Support deflection. Content production. Merchandising and search. Forecasting. Internal tooling. Every one of them is real, every one has a vendor who will demonstrate it beautifully next Tuesday, and there is no natural mechanism inside the business for deciding which of them deserves your team’s finite attention this year.

AI is the loudest source of that problem, not a separate problem. It has never been easier to add another plausible initiative to the list.

Why the usual response doesn’t work

A tools list is the fastest thing to produce and the least useful thing to have.

It answers the question that was asked, which is the trouble with it. Licences get bought, pilots get commissioned, a working group gets a standing invite, and everyone can report activity for three or four quarters. Activity is not nothing. It is just not the same as the P&L moving, and eventually somebody on the board notices the gap between the two.

Think of the office that bought everyone a standing desk. The intent was genuine, the research was real, the invoice was substantial. A year later most of them are at the lowest setting with a monitor on a stack of paper, because the thing that stopped people standing up was never the absence of a desk that goes up. It was the meeting culture, the back-to-back diary, the fact that the room booking system has no concept of a fifteen minute gap. The capability was purchased. The behaviour was never addressed. Nobody was lying, and nothing improved.

Buying capability is a way of looking like you have made a decision without having made one. The brands that get value out of these tools are rarely the ones who bought first. They are the ones who already knew which five things mattered, and could therefore tell in an afternoon whether a given tool moved any of them.

The brands winning with AI didn’t start with AI. They started with clarity.

Proof

Both sides of the table.

Six months embedded at Travel Counsellors on digital experience and AI adoption, which is the operator’s version of this question, asked daily rather than quarterly. Non-executive advisor on technology and AI to a PE-backed consumer brand, which is the board’s version of the same question, asked with a different set of pressures behind it. Retained, both rooms.

The long-form argument sits in Look Again, and the patterns it draws on are published as they are written in Thoughts.

If that isn’t quite it

The diagnostic gets called at four moments.

This is one of them. The other three:

• “The investment has landed.”
• “Growth has flattened and nobody can say exactly why.”
• “We’re being pitched solutions weekly.”

All four run the same four to six weeks and the same three questions. The moment changes where we start looking, not what the work is.