The AI master plan: the management question AI did not change

A business owner knows how to evaluate an investment. What does it cost, what does it return, and how soon. They have been doing it for twenty years without thinking about it.

Then the word AI enters the room, and that discipline disappears.

This is not naivety. The conversation is framed wrong from the start. The vendor talks about what the tool can do. Nobody talks about the line on your financial statements it is supposed to move.

BCG published a number on August 31st, 2026 that sums up the gap: 82 percent of CEOs say they are more optimistic than a year ago about AI returns, and 6 percent of companies see real value in their costs or revenue. BCG attributes the gap to one simple thing: we measure activity instead of measuring financial value.

A master plan does nothing more sophisticated than put the usual question back at the centre.

WHAT A MASTER PLAN ACTUALLY CONTAINS

Three things, and none of them is technological.

One line per initiative, and the financial statement line it touches. By how much, starting when. An initiative that cannot name its line is not set aside because it is bad. It is set aside because we will never know whether it worked.

A measurement taken before starting. Without a baseline at the start, there is nothing to compare at the finish. It is the most frequently forgotten condition and the least expensive one to respect.

An order. Not a ranking of the most promising ideas: a ranking of what fits inside your real capacity for the next quarter. If nobody stops their current work so an initiative can move, it is not prioritised. It is added.

AND A FOURTH THING, USUALLY DISCOVERED TOO LATE

On the night of September 15th to 16th, our own system deleted 56 of its 64 prospecting steps.

Eight sequences, seven steps each. By morning, one step per sequence was left. And the step that survived, every time, was the last one: the closing message, the one that says last email, still relevant. Nine hundred and fifty-three contacts were now scheduled to receive a final reminder as their first contact.

The cause is nothing exotic. An automated call that replaces instead of modifying: it sends the full list of steps, and anything not on the list disappears. The behaviour is documented. The tool worked exactly as designed.

What was missing was not a better tool. It was a decision nobody had made: who is allowed to overwrite what.

That is the question nobody asks before plugging an AI into their systems. We ask whether it will give the right answer. We do not ask what it is allowed to destroy when it gets things wrong.

So a serious master plan contains a fourth column: write permissions. Which system each initiative may read, which it may modify, which it must not touch. And for every write: does it replace, or does it add?

THE SECOND LIST

A master plan produces two lists. Everyone looks at the first one.

The second is the list of things you stop. It is shorter and more profitable, because it frees the one resource a small company actually lacks: the attention of the people who decide. The tool three people use and everyone pays for. The weekly report nobody opens any more. The pilot that never had an end date, that did not fail, but was never evaluated either.

Stopping those three requires no technology and costs nothing. It just does not make a nice announcement.

WHERE TO START

Take your AI initiatives in flight. For each one, three questions.

Which financial statement line it touches, and by how much.

Who stops their current work so it can move.

What breaks if it gets things wrong, and who is allowed to overwrite what.

The ones that answer all three are your plan. The rest are your second list.

And if none of them answers, this is not an AI problem. It is a plan that has not been written yet.

If you want to talk it through: https://calendar.app.google/DKJn1gZgSaRupR4L9