The technology debate is over. The strategy debate hasn’t started.

Most executive conversations about AI are still framed around technology choices: which model, which platform, which integration partner, which proof of concept to fund next. These are real questions. They are also the wrong questions to be asking at the top of the house.

Spend a few hours inside a typical European mid-market or large enterprise AI program and the pattern is consistent. AI is owned by the CTO, sometimes the CIO. It lives inside the technology function as a portfolio of experiments. The conversations that drive it are about budget envelopes, technical feasibility, vendor selection, infrastructure readiness, and data plumbing. Those are legitimate conversations. None of them are strategy conversations.

In most cases there is no center of excellence worth the name, no business-led prioritization framework, no Chief AI Officer with a real mandate. Use cases get picked because a team lead pushed them, because a vendor demoed them, or because they were cheap to try. Value creation, when it is mentioned at all, is invoked at the end of the slide — not at the start of the decision.

What is almost always missing is the only document that actually matters: a sharp, contested, board-level answer to the question of how AI changes what this company is, what it sells, who it competes against, and why a customer would still pick it three years from now.

In the absence of that document, every downstream choice is essentially arbitrary. Pilots get launched and quietly retired. Workflows get “AI-enabled” without anyone asking whether the workflow itself should still exist. Tools get bought because peers bought them. Activity gets measured in proofs of concept and seats deployed, not in margin, share, or repositioning.

Then, eighteen months in, the CEO turns to the CFO and asks where the ROI is, and the answer is some version of: we have a few pilots, we have learned a lot, and we have made parts of the existing business slightly more efficient. Sometimes.

AI is not an innovation project. It is a strategy reset.

The framing error sits at the very top. In a large number of companies I work with, AI has been quietly classified as either an innovation topic or a technology topic — and the two often blur into the same posture. It lives in an innovation lab, a digital factory, an AI guild, an IT roadmap, a CTO’s portfolio of bets. It is treated as a capability the organization experiments with, deploys, secures, and gradually industrializes. What it is almost never treated as is what it actually is: a business upheaval.

That model worked reasonably well for previous waves: cloud migration, mobile, data, even early machine learning. None of them, in the end, required a CEO to rewrite the company’s value proposition. They were horizontal capabilities that could be absorbed by the existing operating model.

Generative and agentic AI are not that. They are not a capability you bolt onto an unchanged business. They alter, simultaneously and at different speeds, four things every strategic plan is built on:

• The value proposition. What the customer is actually paying for shifts when knowledge work, advice, design, code, and decisioning can be produced at a fraction of the current cost and time.

• The cost structure. The unit economics of services-heavy businesses, in particular, are about to be rewritten by competitors willing to take the gross margin hit first.

• The competitive perimeter. Adjacent players, software vendors, native AI entrants, and your own clients can now credibly attack parts of your value chain they previously couldn’t.

• The sources of advantage. Scale, distribution, brand, and data moats are being repriced. Some get stronger. Several quietly evaporate.

You cannot delegate that conversation. Not to a CTO, not to a Chief AI Officer, not to an external consultant, and certainly not to a steering committee whose mandate is to “accelerate adoption.” These are first-order strategic questions, and they sit on the CEO’s desk, with the executive committee and the board as co-signatories.

Delegating AI to your CTO is not a governance choice. It is a strategy choice — and a bad one.

When AI is owned at the CTO level, the conversation defaults to enablement: what do we deploy, to whom, at what cost, with what governance. Those are necessary questions. They are not sufficient ones. They cannot, by construction, surface the question of whether the entire business should be reshaped.

Why the ROI isn’t coming

Once you accept that the upstream framing is missing, the ROI problem becomes almost mechanical. There are four ways an AI investment can return capital:

• Cost: structural reduction of the cost base, not marginal task acceleration.

• Revenue: new offers, new pricing models, new addressable markets.

• Quality and risk: fewer errors, faster cycle times, better decisions, reduced exposure.

• Strategic optionality: positions and capabilities that pay off in scenarios competitors haven’t priced yet.

Three of those four require a strategic decision that an enablement program cannot generate on its own. You can roll out Copilot to ten thousand seats and not move any of them in a measurable way, because nothing about the business model has been asked to change.

The companies starting to show real returns are not the ones with the highest adoption metrics. They are the ones where someone in the C-suite, eighteen to twenty-four months ago, sat down and asked a different question: not “how do we adopt AI?” but “what does our business look like if we assume AI is true?” Different answer. Different roadmap. Different ROI.

Everyone else is running a very expensive productivity experiment and calling it a transformation.

The contrarian read

There is a pattern in how this is currently being narrated in the market, and it deserves to be named.

When pilots underdeliver, the story line in most boardrooms is: the technology isn’t mature yet, the models will get better, costs will come down, the use cases will sharpen, give it another year. That story is comforting because it externalizes the problem. It implies that ROI will arrive on its own, carried by the next model release.

It won’t. Better models will not fix a missing strategy. They will, if anything, make the gap more expensive: more capacity bought, more workflows automated around an unchanged business, more capex sunk into a posture that needed to be redesigned, not optimized.

The honest read is harder. In the cohort of companies that will look back on this period and conclude that AI generated meaningful value, almost all of them will share one feature: a CEO who treated this as a strategy problem from the start, owned it personally, and was willing to put the value proposition, the operating model, and the competitive positioning back on the table. Not as a workshop. As a decision.

Everyone else will keep finding reasons — the models, the data, the change management, the regulator, the talent gap — and will keep being right about each one in isolation, and wrong about the whole.

Better models will not save a missing strategy. They will make its absence more expensive.

What a serious answer looks like

This is not a call for another transformation program. It is the opposite. Most companies do not need more AI activity. They need a smaller number of much harder decisions, taken at the right level, in the right sequence.

In the pieces that follow this one, I’ll lay out the operating frame I use with executive teams and boards to do exactly that. Three deep dives, designed to be read in order:

• The four dominant AI strategies. Not a maturity model. A strategic choice. What kind of AI company you are choosing to be — and what you are choosing not to be — with the trade-offs each posture forces.

• The seven critical success factors of an AI strategy. What separates the executive teams that compound returns from the ones that compound spend. Almost none of them are about technology.

• The four components of AI ROI. How to actually measure it, what to count, what to stop counting, and how to talk about it with a board that is rightly running out of patience.

Taken together, that is the conversation the C-suite should have been having from the beginning. Most haven’t. There is still time — not much, but some — to have it now.

CEO Move

Before the next AI steering committee, before the next budget cycle, before the next vendor pitch, take an hour with no slides and answer four questions on paper:

• If AI delivers on its current trajectory, what part of our value proposition is still defensible in three years — and what isn’t?

• Which of our competitors, including the ones we don’t take seriously yet, are most likely to use AI to attack us, and on which margin?

• If we were starting this company today, knowing what we know, what would the operating model look like — and how far is that from where we are?

• Who, by name, owns the answer to the three questions above? If the name on that list is not yours, the strategy problem is already the answer to the ROI question.

If you can’t answer those four, no model, no platform, no enablement program is going to save the investment. The ROI gap is not a technology gap. It is the visible part of a strategy that was never written down.