Module 04 Synthesis — The Executive AI Strategy

In 2025 Boston Consulting Group asked 1,250 senior executives who make decisions about AI how much value their companies were getting from it. Before reading on, make a guess: what share said they were getting no material value at all, despite substantial investment?

The answer was 60 percent. Only 5 percent were achieving value at scale¹.

BCG’s description of what separates the two groups reads like a definition of coherence. The firms getting value have top management that “translates overall business goals into a multiyear, fully funded AI vision with an explicit execution roadmap”, and their value comes “not from AI pilots or isolated use cases” but from reshaping core workflows end to end. Among the laggards, top management “fails to articulate any clear value ambition” and delegates AI downward¹. The survey is self-reported and run by a consulting firm, so it shows an association rather than a cause. The association points one way: AI pays where the choices about it hold together.

The pattern is older than AI. In a 2011 survey of more than 1,800 executives, Booz & Company found that 49 percent said their company had no list of strategic priorities, and 54 percent said its capabilities did not reinforce one another². Those firms were not short of activity either. They had initiatives, budgets and teams. What they lacked was the thing that turns activity into direction: a set of choices that fit.

One strategy, not fourteen answers

Each chapter of this module answered one question well. The risk now is to treat the module as a checklist: an ambition statement here, a sourcing decision there, a platform, a data plan, a talent plan, a decision-rights chart. Every item can be sound and the whole can still fail, because the items were chosen separately, by different owners on different calendars, and pull in different directions.

A. G. Lafley and Roger Martin, writing from a decade of strategy work at Procter & Gamble, define strategy as an integrated set of five choices: a winning aspiration, where to play, how to win, the capabilities that must be in place, and the management systems that support them. The weight falls on the word integrated. The choices must reinforce one another, or the organization ends up with several strategies arguing instead of one³. Richard Rumelt’s kernel of good strategy makes the same demand: a diagnosis, a guiding policy and actions that are coherent with each other⁴.

So the claim of this synthesis is simple. An executive AI strategy is one chain of choices that runs from the business problem to a defensible advantage, and each link has to fit the links beside it.

The module in four layers

Module 04 builds that chain in four layers, each resting on the one beneath it.

Read from the bottom, the layers make one argument. Know what the business must achieve and how much AI should change it. Choose where to invest, what to own and how to organize the work. Build the data, people and ownership to execute. Only then claim an advantage, and be honest about which uses of AI merely keep pace with rivals and which could set the organization apart. The layers are a map for study. A leadership team needs something it can ask in a meeting.

Five links make the strategy chain

Five links do the job, each a question asked in order: why, where, what, how and win. Call them the five links of the strategy chain, not to be confused with the five decision-rights questions in AI Decision Rights and Accountability.

Why asks what business problem AI must help solve and how far AI should change the business. Where asks which few opportunities deserve investment, and which attractive ideas will not get any. What asks which capabilities and assets the organization must own and which it can buy or reach through a partner. How asks how the work will run: the shared platform, who does what, and who decides. Win asks what, if anything, will make the organization’s use of AI hard for a competitor to reproduce.

The links are not new inventions. They line up closely with the strategy cascade Lafley and Martin describe, which is a reassurance: an AI strategy is a business strategy with AI in it, not a separate discipline.

One difference in order is deliberate. Lafley and Martin place “how to win” beside “where to play”, at the heart of the cascade. In AI it pays to ask it last as well as first, because the honest answer often changes once the capabilities are on the table. A claim of advantage made before anyone has asked what must be owned is usually a hope.

Coherence: choices that argue with each other

The most common failure in an AI strategy is not a bad choice. It is two reasonable choices that contradict each other. Michael Porter’s argument in What Is Strategy? is that advantage comes from fit among activities, where each one makes the others more valuable, rather than from any single activity done well⁵. Fit is also what makes a position hard to copy: a rival can buy any one of your choices, but not the way they lean on each other.

These contradictions are rarely anyone’s mistake. Each choice is made by a different owner, on a different calendar: procurement signs the platform contract, the business units launch pilots, HR sets the hiring plan, legal writes the customer terms. Each decision is defensible in its own meeting. The contradiction appears only when someone reads them side by side, which is the job of the executive who owns the strategy. If the advantage is meant to come from what you know, you cannot outsource the knowing. If speed matters, authority has to sit where the knowledge is, scaled to the stakes. And an ambition the organization cannot staff is an announcement, not a choice.

Marco Iansiti and Karim Lakhani make the same point at the scale of the whole firm: companies that compete on AI do not bolt models onto an unchanged business, they rebuild the operating architecture around data and learning⁶. Coherence is that rebuild, seen from the strategy side.

A practical test follows. Take any two choices in your strategy and ask: if we carry out both in full, does one make the other easier or harder? Where the answer is “harder”, the strategy has a seam that will open under pressure.

An arch, not a pile of stones

Think of a stone arch. No single stone holds anything up. Each stays in place only because the stones on either side press against it, and the keystone at the top locks them all. A pile of excellent stones next to the arch holds up nothing at all, however good each stone is.

A list of AI initiatives is the pile. Each pilot may be well cut. A strategy is the arch: every choice leans on the others, and the keystone is the why, the business problem that tells each other choice which way to lean. Remove it and the rest come down, which is exactly what happens when an AI program loses its sponsor’s reason for existing.

The metaphor breaks in one useful place. Once built, an arch is meant never to move. A strategy has to be rebuilt stone by stone as evidence arrives, while it is still carrying weight. Realized strategy always mixes what was intended with what emerged along the way⁷. A review on a fixed rhythm, with the authority to move money, is how the arch is re-laid without falling down.

Writing it down: five sentences and a stop

A strategy that holds together can be written briefly. A useful discipline is to write the AI strategy as five sentences, one per link, and then add a sixth that begins “we will stop”. If any sentence is vague, the thinking behind it is vague too.

The sixth sentence matters most. As Porter put it, the essence of strategy is choosing what not to do⁵. A document with no “stop” in it has not yet made a choice that disappointed anyone, which usually means it has not made a choice at all.

Story: one day at a dairy cooperative

The chief strategy officer of a farmer-owned dairy cooperative in northern Europe starts her day at half past six, when the first milk tankers leave for member farms. Every pick-up is sampled, and the cooperative’s laboratory tests each sample for fat, protein and cell counts. A message from the chair is waiting: the board wants the AI strategy at Friday’s meeting.

By nine her team has listed every AI initiative in the group. There are twenty-three: a tanker-route optimizer, a tool that drafts responses to retailer tenders, generated copy for the cheese brand, demand forecasting, contract review, a herd-health alert trial run with one milking-robot maker, and more. Each has a sponsor. None is tied to the cooperative’s strategy, which is to move members’ milk out of commodity powder and into specialty cheese and premium contracts that pay more.

At lunch she sits with a farmer on the board. He is blunt about one thing. The data from his milking robot is his, and members will not hand it to anyone who uses it without asking. That conversation changes the what question. The cooperative cannot own its members’ farm data. What it can earn is their consent, and what it already owns outright is something no rival has: a daily laboratory record of the milk from every member farm.

By three she has rewritten the plan as choices. Why: lift the margin on specialty products. Where: two value pools that stand on the same asset, routing milk of the right composition to the right cheese plant, and farm advice on herd health that members can choose to receive. What: own the laboratory data and the evaluation of every model; buy route optimization and writing tools as ordinary productivity software. How: one shared data platform, plant managers deciding schedules within agreed limits, and the farmer and the vet deciding every treatment. Win: members’ trust, and a record nobody else can buy.

Then she tests the fit, one pair of choices at a time. The herd-health trial gives the robot maker exclusive use of the alerts data, which argues directly with the win sentence, so it is renegotiated or stopped. A rule that every AI project needs board approval argues with the plant teams’ need for speed, so authority is delegated by stakes. Nine pilots that sit outside both value pools stop.

The fit test has a price, and she cannot make it disappear. One of the nine is the tender-drafting tool, the pet project of the commercial director, who sits on the executive team and says so. The robot maker may walk away rather than give up exclusivity, and then the herd-health pool has no live data until members consent, which could take a season or more. She decides not to hide either cost. The board paper names the stopped pilots and their sponsors, and it says plainly that the second value pool will be late.

At half past six in the evening she rehearses the board answer. It takes five minutes and ends with the sentence that starts “we will stop”. Nothing new was bought that day. The strategy came from making the existing pieces lean on each other, and from saying out loud who loses.

What this means for leaders

The module’s lessons reduce to a short practice. Start from the business problem, not the technology, and let it decide which way every other choice leans. Work through the five links in order, and expect the answer to win to change once you have answered what. Test every pair of choices for fit, because the seams between sensible choices are where AI strategies quietly fail. Write the result in five sentences and a stop, treat the stop as the proof that choices were made, and name the sponsors who lose. Finally, keep the strategy alive: review it on a rhythm and move money as evidence arrives.

Check yourself

1. A portfolio of well-run AI pilots adds up to an AI strategy.MythFact
2. If each Module 04 choice is made well on its own, the strategy will be sound.MythFact
3. Lafley and Martin define strategy as an integrated set of five choices.MythFact
4. In BCG’s 2025 survey, most companies reported getting material value from AI.MythFact
5. It is safe to claim the win before deciding what the organization must own.MythFact
6. Stopping an AI initiative is a sign that the strategy has failed.MythFact

Reflection: find the seam

What comes next

Module 04 began with a claim that an AI strategy is a short set of hard choices and ends with the shape those choices take together: a business problem at the keystone, a few value pools, a clear line between what is owned and what is bought, a way of running and deciding, and an honest view of where advantage can come from. A strategy, though, creates no value until it is applied to real work. Module 05 turns from choosing to applying, and its first chapter, The Enterprise AI Use-Case Landscape, maps where AI is actually being used across the enterprise, function by function.