Before you buy an AI strategy, check which problem you have

Requests for an AI strategy tend to arrive in the same shape. The board has asked, a competitor has announced something, a budget line exists, and somebody has been told to come back with a plan by the end of the quarter.

Sometimes a strategy engagement is exactly the right purchase. Often it is the right answer to a different problem from the one the company actually has, and it produces an accurate document that changes nothing. Telling the two apart takes about an hour of honest conversation, and it is worth having before anybody signs anything.

Four situations that all get called “we need an AI strategy”

There is no agreement on where to start. Budget and mandate exist, five departments each have a favourite idea, and every meeting reopens the same argument. This is what strategy work is for: a short ranked list, reasoning done in the open, and the rejected ideas written down so they stop coming back.

You already know what to build. A process with a clear owner, data that exists and a cost everybody agrees on. A strategy engagement here is a delay with a fee attached. Go to engineering, and spend a day agreeing how you will measure the result.

The decision has been made and needs a document. Strategy work used as justification tests nothing, and the people interviewed for it can tell. If the decision is made, own it openly and spend the money on doing it well.

There is no capacity to change anything. An ERP migration, a reorganisation and two other programmes are already running on the same people. A strategy written now will be correct and shelved. The first move is to stop something, which is a management decision rather than a consulting deliverable.

Infographic

Is this a job for strategy at all?

Capacity to changeNo capacity

AI strategyBudget and mandate exist; agreement on where to start does not.

Go straight to buildYou know what, who and on which data. Strategy only delays.

Fix the basics firstNeither a target nor people to pursue it. The plan gets shelved.

Make room firstYou know what, but not with whom. Stop something else.

Unclear targetClear target

What it should cost you, apart from the fee

The invoice is the visible part. The time is the part that decides whether the result is any good.

Elapsed time: weeks, not months. Two to four weeks is enough to frame the problem for most scopes. Anything longer is a programme, and it should be bought as one, with a different contract and different expectations.

A handful of people for a few hours each. Whoever decides, whoever executes and whoever owns the data, typically two to six people. Skipping the people who execute is the most common way to save time and the most expensive, because the plan then describes a company that does not exist.

Somebody to pull the evidence. Actual volumes, cycle times and error rates, not estimates. This is usually the slowest part of the work and the one most worth doing, because perception and measurement disagree more often than they agree.

The sponsor, twice. Once at the start to agree scope, and once at the end to make the decision. A sponsor who cannot attend the final session turns the document into a recommendation waiting for a meeting.

The fee itself should be fixed and agreed before the work starts. Framing billed by the day carries an incentive to keep framing, and nobody should have to manage a supplier around that.

What a good result looks like

A short ranked list — three to five candidates, not thirty — each with a named owner, a cost to build, a cost to run and a measurable definition of success. A written record of what was rejected and why. And the strongest argument for doing nothing, because a recommendation that has never faced its best objection has not been tested.

There is a simple check on quality: could you hand the document to a different supplier, or to your own team, and have them start? A strategy that only works if its author builds it is a sales document.

A good result can also say that the answer is not AI. A process fix and a data cleanup are a legitimate outcome, and often a cheaper one.

The hour before

Before buying anything, answer three questions internally. Which decision or process costs us the most today? Who would personally own a change to it? What would we stop to make room?

If the first two have answers and the third gets silence, the problem is capacity, and a strategy will not solve it. If none of them has an answer, that is exactly the situation strategy work is built for.

That is how we scope an AI strategy engagement: a fixed price, a few weeks, a short list with owners and numbers, and the honest version of the answer even when it is not the one that was commissioned.