Last Decade’s Business Case, This Year’s AI Strategy
Faced with something genuinely new and poorly understood, an organisation reaches instinctively for the shape of the last thing it understood — and the last thing it understood was cloud.
The Document That Gave Itself Away
The giveaway is usually somewhere around slide twelve.
By then the board pack titled Enterprise AI Strategy has moved through the ambition (“become an AI-led organisation”), the market context (a chart of investment forecasts with a steep line), and the guiding principles. And then it arrives at the workstreams — and if you have been in this profession long enough, you feel a flicker of recognition. Consolidate the data estate. Complete the migration to the cloud platform. Stand up a centre of excellence. Upskill the workforce. Put governance around it. You have read this document before. Three or four years ago it was called the cloud strategy, and before that the digital strategy, and the only material edit has been the find-and-replace in the title.
What has caught our attention over the last year — as the wave that began with a chatbot late in 2022 has washed into every boardroom — is how many of the AI strategies now circulating are cloud strategies with the serial numbers filed off.
This is worth saying out loud, because it is happening quietly and it is easy to miss. Nobody sets out to recycle. The recycling is not laziness, and it is not stupidity; it is structural, which is why it is so hard to see from the inside. The same executive commissions the strategy. The same advisory firm writes it, drawing on the same reference architecture that served it well for a decade. The same procurement machinery prefers a platform line-item to an ambiguous question about the business. Faced with something genuinely new and poorly understood, an organisation reaches instinctively for the shape of the last thing it understood — and the last thing it understood was cloud.
A cloud programme asks how we run our existing work more cheaply and flexibly. An AI ambition should ask a harder question: which decisions, and which work, are different now that a machine can produce plausible language and prediction on demand? The recycled strategy answers the first question and quietly files the second under “phase two”.
Consider a composite that will be familiar. A financial-services group publishes a five-workstream “AI strategy” this year. Workstream one: data platform consolidation. Two: finish the cloud migration. Three: an AI centre of excellence. Four: a skills academy. Five: a responsible-AI governance framework. Of the five, exactly one — the third — contains anything that would not have appeared, near-verbatim, in the group’s 2019 cloud programme. Four-fifths of the “AI strategy” is the completion plan for the previous strategy, rebadged. The budget follows the same split: the overwhelming majority flows to platform and migration, and the sliver left over funds a proof-of-concept backlog nobody has prioritised against a single business decision.
There is a serious objection to all this, and it deserves a straight answer. The foundations genuinely do matter. You cannot retrieve what you have not governed, and a language model pointed at an ungoverned data estate produces confident nonsense faster than before. The people writing these strategies are not wrong that the data and platform work has to happen. Where they go wrong is subtler: they let the foundation stand in for the question. “Get the data ready” becomes the whole plan, and readiness is treated as a destination rather than a means. The cloud migration was allowed to be its own justification once — capability for its own sake — and the same permission is now being extended to data-for-AI. The result is a strategy that can be delivered in full without a single decision in the business ever being made differently.
The tell is simple, and any board can apply it this quarter. Take the AI strategy and strike out every sentence that would have been equally true in the cloud programme. If what remains is a centre of excellence and a list of tools, there is no AI strategy in the document — there is a platform plan wearing this year’s language.
What actually distinguishes an AI ambition from its cloud predecessor is not the size of the platform but the presence of a decision it changes: a specific judgement, currently made by people in a particular role, that will be made differently — faster, at greater scale, with a machine in the loop — and the honest work of saying what “differently” means, who is accountable when the machine is wrong, and how the organisation will know. That is uncomfortable, specific, and unglamorous. It is also the only part that is genuinely new. Everything else, we have built before.