Ninety-three percent. That is how many AI initiatives stall before the finish line. Not because the technology fails. Because the strategy is missing.
What most organisations call an AI strategy is a list of tools they want to try. Roll out Copilot. Build a chatbot. "Something with automation." These are answers to a question nobody has asked. Which problem are we solving? Where is the friction? What can we do now that we could not do last month?
Without those questions, an AI strategy is a shopping list without a recipe.
What an AI strategy actually is
An AI strategy is an explicit decision about where, as an organisation, you focus your capacity to learn. Which domain, which team, which process. With what resources, within what timeframe, and how you measure whether it works.
That sounds simple. In practice, it is almost always absent. Organisations start broad ("we are going to apply AI across the entire organisation"), get fragmented, and end up with six separate pilots that nobody makes a decision about.
The core of a working AI strategy is focus. Pick one fight. Go deep on it. Every extra scope halves your learning capacity.
Three elements that matter
A bounded playing field. "Customer service" is too broad. "First-line email triage for product complaints" is exactly right. The playing field fits on one page. If it does not fit on one page, the scope is too wide.
Measurable criteria defined upfront. How will you know it works? Define that before you start building. And always measure in pairs: speed alongside quality, cost savings alongside employee satisfaction. One-sided metrics lead to one-sided conclusions.
A forced decision. At the end of the cycle, you force a choice: scale, adjust, change direction, or stop. All four are good outcomes. "We're keeping an eye on it" is not. What that looks like in practice is described in AI strategy in ninety days.
Where it goes wrong
Two patterns we encounter in almost every organisation.
The first: starting with the tool. "We need to do something with agents." "Have you looked at Copilot yet?" The technology is central. The problem comes later. Or never. That is Solutioneering, the fastest route to an expensive project that delivers nothing.
The second: no owner. An AI strategy without someone who has the authority to stop is a strategy without brakes. You need a sponsor who can say "no." To the CEO, to the vendor, to the enthusiasm when it is heading in the wrong direction.
The capabilities shift
Most AI strategies are about efficiency. Speed up processes, reduce costs, save headcount. That is half the story.
The other half: AI makes things possible that were impossible last month. A retailer that can personalise its range in real time for each customer. A manufacturer that can turn around every quote within an hour instead of three days. A healthcare organisation that recognises patterns in patient data that no human would notice.
These are not process improvements. They are new capabilities. The interesting AI strategy is about that shift: what can we do now that we could not do yesterday?
The sequence
Start with the friction. Inventory where people are already using AI tools on their own initiative. Choose the domain with the most pain and the most potential. Build something small in two weeks. Measure it. Decide.
Repeat. Each cycle builds on the previous one. After three cycles, you have an AI strategy that has grown organically from evidence, rather than being designed top-down in a meeting room.
Ninety-three percent never reach production. The seven percent that do started small, with focus, with an owner, and with the willingness to stop when the evidence was not there.