From AI strategy
to real value

How do you turn an AI strategy into measurable results?

An AI strategy becomes executable with three translation steps: turning ambition into measurable business outcomes, assigning one owner with budget and a mandate per use case, and measuring value at the level it was promised at. Most AI strategies don't fail on the strategy itself, but on translating it. A portfolio of four use cases with owners, measurement points and stop criteria delivers more than a roadmap with twenty initiatives nobody is accountable for.

Does ownership of a use case from the AI strategy belong with the AI lead or the data team?

No. Placing ownership with the AI lead or the data team is a pitfall: they can build and advise, but they don't sit on the P&L where the value has to land. The owner of a use case from the AI strategy belongs in the business.

Do usage figures prove that an AI strategy delivers value?

No. Usage statistics are useful as an early indicator, but not proof of value for an AI strategy. A business case formulated in pounds or hours ultimately needs reporting in pounds or hours too, even when that's harder to measure.

Is a baseline measurement needed to prove the result of an AI strategy?

Almost always: measuring the business result of an AI strategy almost always requires a baseline measurement before the start. That baseline measurement is two days' work, and without it, nothing can be proven a year later.

Source: article by Dave Willemstein. Updated: October 2026.

Where it goes wrong between strategy and results

I rarely see a bad AI strategy. What I do often see: a perfectly good strategy that, nine months later, is still exactly the same document, with the same ambitions and not a single result delivered. Not through unwillingness, but because three translation steps are missing that nobody has explicitly assigned.

Those three translation steps aren't a methodology. They're three questions you need to be able to answer before a strategy becomes executable.

Step 1: Translate ambition into business outcomes

Most AI strategies are written in technology language: "we're deploying generative AI for customer contact", "we're building a data-driven decision culture". Those are directions, not outcomes. You can't turn them into a quarterly target, and you can't steer by them.

The translation that works looks like this:

• From: "using AI for customer contact", to: "average handling time for first-line queries down from 8 to 5 minutes, with customer satisfaction unchanged."
• From: "predictive maintenance", to: "unplanned downtime on lines 3 and 4 down 20% within twelve months."
• From: "increasing knowledge-worker productivity", to: "two hours per consultant per week freed up, redeployed to billable work."

The difference looks cosmetic. It isn't. The right-hand column forces you to decide who's accountable, what data you need to measure it, and when you'll conclude it isn't working. The left-hand column leaves all those questions open, and that is exactly why those strategies stall.

The pitfall here: formulating outcomes so ambitious that nobody dares to claim them. A goal nobody wants to attach their name to isn't a goal.

Step 2: Assign ownership with budget and a mandate

This is the step that gets skipped most often, usually with the best intentions. The use case gets "assigned to a working group" or "picked up jointly by IT and the business". In practice, that means: when it gets difficult, everyone looks at each other.

Ownership that works has three characteristics:

• One name. Not a department, not a duo. One person who explains in the steering group why it is, or isn't, moving forward.
• Budget in their own line. If the owner has to go through someone else for every pound, that someone else is the real owner.
• A mandate to stop. This is almost always forgotten. An owner who is only allowed to continue is an executor.

A practical test: can you name the owner of every use case on your roadmap within ten seconds? If not, that use case doesn't have one.

The pitfall here: placing ownership with the AI lead or the data team. They can build and advise, but they aren't the ones sitting on the P&L where the value has to land. The owner belongs in the business, even when that's uncomfortable because that person has no AI background.

Step 3: Measure at the level you promised

This is where most of the quiet disappointment happens. The business case promised £400,000 in savings. The reporting shows: 1,200 active users, 87% satisfaction, 40,000 documents processed. All true, all positive, and none of it an answer to the question that was asked.

Usage statistics are useful as an early indicator. They are not proof of value. If your business case was formulated in pounds or hours, you ultimately need to report in pounds or hours too, even when that's harder to measure.

Three measurement layers that work together:

• Adoption (weekly active users, usage per scenario): tells you whether people are using it.
• Process performance (lead time, error rate, rework): tells you whether it's working.
• Business outcome (cost, revenue, capacity): tells you whether it delivered anything.

The third layer almost always requires a baseline measurement you should have taken before you started. Take that baseline measurement. It's two days' work, and without those two days you can't prove anything a year later.

The pitfall here: only measuring what happened to already be on a dashboard. That's how you end up with a report full of usage figures.

What this delivers in practice

These three steps aren't complicated. They're uncomfortable. They force you to shrink ambitions down to something someone is willing to put their name to, and to establish in advance when you would conclude that it hasn't worked.

That's exactly why they work. A portfolio of four use cases with owners, measurement points and stop criteria delivers more than a roadmap with twenty initiatives everyone is enthusiastic about and nobody is accountable for.