The AI Strategy That Was Really a Cloud Strategy
A strategy that can be produced by find-and-replace was never a strategy; it was a procurement plan wearing the vocabulary of ambition.
The deck I had already read
The meeting was convened to approve an AI strategy, and the strategy turned out to be a deck of forty slides. By the eighth slide I had the uncomfortable sense of having read it before; by the twelfth I knew where. It was the cloud strategy the same organisation had signed off a few years earlier, carried forward almost whole. The architecture diagrams were the same, redrawn in a warmer palette. The governance model was the same, relabelled. The migration roadmap was the same, its destination renamed. Somebody had performed the corporate equivalent of find-and-replace — cloud had become AI, data lake had stayed a data lake, centre of excellence had survived untouched — and the result was being presented as a plan for the next decade.
I do not tell this story to mock the people in the room. They were serious and capable, and the pressure they were under was real. A general-purpose chatbot had arrived on every desk over the winter, the board had asked the only question boards ask when a technology reaches the front pages — what is our strategy for this? — and a document had to exist by the next meeting. What they produced was the fastest honest thing available: the last big technology plan, refreshed. The trouble is not that they were careless. The trouble is that a strategy you can assemble by find-and-replace was probably not a strategy the first time either.
Strategy recycling
I have watched this reflex play out across enough technology waves now to treat it as a pattern rather than an accident. The enterprise resource planning wave, the web wave, the “big data” wave, the cloud wave — each arrived wrapped in the same organisational response. Buy the platform. Stand up a centre of excellence. Consolidate the data. Name a handful of use cases. Draw a roadmap with three horizons. Call it the strategy. The nouns change with each cycle; the deck does not. Generative AI is simply the latest noun.
There is a reliable way to tell a recycled strategy from a real one, and it has nothing to do with the quality of the slides.
The reliable tell of a recycled strategy is this: it describes at length what the organisation will build and buy, and is almost silent on what the organisation will decide differently once it has.
A genuine strategy is a set of choices about where to compete and how to win, and choices are painful, because they close doors. A recycled strategy is a shopping list, and shopping lists are pleasant, because everything stays open. When the “AI strategy” spends thirty slides on data platforms, target architectures and operating models, and three slides on use cases described in a sentence each, what you are looking at is not a strategy for artificial intelligence. It is a procurement plan that has borrowed the vocabulary of ambition.
Why the reflex is so hard to resist
It would be easy to put this down to laziness, but that lets us off too lightly. The recycling of strategy is over-determined: several forces push in the same direction, and each is rational from where the actor stands.
- Infrastructure is buyable; judgement is not. You can sign a contract for a data platform on Friday and point to progress on Monday. You cannot sign a contract for better decisions. Faced with a demand to do something, organisations gravitate to the part of the problem that money can visibly move, and infrastructure is that part.
- Vendors sell platforms, and a strategy tends to inherit its supplier’s frame. The firms best resourced to help you write an AI strategy are, overwhelmingly, the firms selling the underlying platforms. Their reference architectures are generous and genuinely useful, and they quietly answer the question what should we buy? rather than what should we change? Borrow the frame and you inherit the conclusion.
- Capability is mistaken for strategy. “We will be able to deploy models against our data” is a statement of capability. It feels like a strategy because it is ambitious and expensive, but it commits to nothing. Almost every organisation in a sector can say the identical sentence, which is the surest sign that it decides nothing.
- Migrations have finish lines; changes in judgement do not. A cloud migration can be declared complete: the servers are gone, the estate has moved. A change in how an organisation actually makes decisions has no such clean edge, no ribbon to cut. Under pressure to show completion, we choose the work that can be finished over the work that matters.
None of these forces is foolish. Together they produce a strategy that is comfortable to write, easy to approve, and empty of choices.
The strongest version of the objection
The obvious retort — and it is a good one — is that the foundation genuinely matters. You cannot do serious machine learning on a fragmented, ungoverned, on-premise estate held together by overnight batch jobs. The cloud migration, the consolidation, the governance model: these are not distractions from an AI strategy, the argument runs, they are its precondition. An organisation that used the arrival of generative AI as the reason to finally fix its data foundations did something sensible, not something fraudulent.
We should grant all of this, and it is exactly where the confusion lives. The foundation is necessary. It is simply not sufficient, and it is not a strategy.
“Cloud was the foundation. The mistake was to hand over the blueprint, admire the poured concrete, and call the foundation the building.”
Necessary is not the same as sufficient, and almost the whole art of strategy lives in that gap. A firm can complete every slide of the recycled deck — migrate the estate, stand up the platform, hire the team — and still have changed nothing about how it competes, because none of that work ever reached a decision. The foundation was real. The building was never designed.
The question a strategy is supposed to answer
If the recycled deck answers what will we be able to do, a real strategy answers a harder and more specific question.
“The question a strategy must answer is not what we will be able to do, but what we will decide differently once we can.”
Put that question to a recycled deck and it falls silent, because the honest answer is nothing yet. Put it to a real one and it becomes concrete and uncomfortable at once: which decisions, made by whom, will change; where judgement moves from a person to a model, or from a model back to a person; what must be true not only of our data but of our people’s willingness to act on a machine’s recommendation when it contradicts their instinct.
Let me make this concrete, because the point dies in the abstract. A financial-services organisation I am thinking of — the details composited, the shape true to life — spent a little over two years and something close to fourteen million pounds executing exactly the recycled deck. At the end they had, precisely as promised, a consolidated data estate, a governed platform, and a capable team. They also had one model in production: a customer-churn predictor. It performed well on its own terms. It was also ignored, because it flagged at-risk customers a fortnight after the retention team’s own instinct had already written them off, and nobody had been given either the authority or the incentive to act on it any earlier. The strategy had delivered every artefact on its roadmap and touched not a single decision. The model worked. The organisation did not.
Generative AI makes this failure mode easier to fall into, not harder. It is now trivial to buy access to a capable model, point it at a pile of documents, and demonstrate something impressive within a fortnight. The demonstration is real; the temptation is to mistake the demonstration for the strategy. A pilot that dazzles a steering committee and changes no operating decision is the churn predictor again, in a more articulate disguise.
A test you can apply this week
It is tempting to reach for grand definitions of strategy; a blunt test is more useful, and this one survives contact with a real steering meeting.
- Take your AI strategy and your last cloud or digital strategy, and read them side by side.
- Mark every sentence that would still be true if you swapped the technology noun back.
- If most of the document survives the swap, you are not holding an AI strategy. You are holding a procurement plan with a fresh cover.
Run this exercise honestly and the proportion that survives is sobering — comfortably more than half of the sentences, in the cases we see, are the previous plan in new clothes. That is not a reason for cynicism about the technology, which is real and consequential. It is a reason for suspicion of the deck.
A strategy that can be produced by find-and-replace was never a strategy; it was a procurement plan wearing the vocabulary of ambition. The foundations are worth building, and the platforms are worth buying. But the strategy is the part the recycled deck always omits: the decisions we will make differently once the foundation is poured — and the harder admission that pouring concrete has never, in any wave I have lived through, been the same thing as knowing what to build.