Why most AI strategies fail in the first 90 days.
Most AI programmes don't fail because the technology is wrong. They fail because the first quarter is spent answering the wrong question.
Ask a boardroom what their AI strategy is, and you will usually hear a list of tools. A copilot here, a chatbot there, a model someone is "exploring" in a corner of the business. None of it adds up to a strategy — it adds up to a shopping list. And a shopping list is exactly the thing that quietly stalls around day ninety.
We see the same arc again and again. Weeks one to four are euphoric: pilots spin up, a demo wows the leadership team, everyone agrees the future has arrived. By week eight the momentum thins. The pilot works, but nobody can say what it changed. By week twelve the project is "paused for prioritisation" — which is the polite way of saying it has been quietly shelved.
The pattern
The problem was never the model. It was the question.
The teams that struggle start with capability: what can this technology do? The teams that succeed start with consequence: what decision do we want to make better, faster, or more often? It sounds like a small reframe. In practice it changes everything downstream — the data you collect, the people you involve, and the way you measure whether any of it worked.
When a programme is anchored to a decision, the path forward is obvious. You know which data matters because it is the data that informs that decision. You know who needs to be in the room because they are the people who own the outcome. And you know what success looks like, because the decision either got better or it didn't.
What actually goes wrong
Three failures that compound.
Across the engagements we run, the first ninety days tend to break in three predictable places. None of them are technical. All of them are avoidable.
- No owner for the outcome. The model has an owner. The decision it's meant to improve does not. Without someone accountable for the consequence, the work drifts toward whatever is easiest to ship.
- Data that describes the past, not the choice. Teams pour energy into cleaning historical data before anyone has asked what question it needs to answer. The result is a beautifully organised archive that informs nothing.
- Success measured in activity, not movement. "We deployed three models" is activity. "We cut quote turnaround from four days to four hours" is movement. Only one of those survives a budget review.
The fix
Start narrow. Earn the right to scale.
The most durable AI strategies we've helped build in the first quarter were almost embarrassingly small in scope. One decision. One owner. One clear before-and-after. The ambition is never small — but the starting surface is, because a narrow start is the only kind you can actually prove.
Pick a decision that is made often enough to matter and is currently made on instinct or incomplete information. Instrument it. Improve it. Show the movement in numbers a CFO would recognise. That single, defensible win is what buys you permission — and budget — to do the next one. Scale is something you earn, not something you announce.
The companies pulling ahead aren't the ones with the most models. They're the ones who turned context into clarity on a decision that mattered, then did it again. That's the whole game. Everything else is a demo.
An AI strategy isn't a list of tools you've adopted. It's a list of decisions you've made sharper.
Before you greenlight an AI initiative, write the sentence: "In ninety days, this decision will be measurably better, and this person will be able to prove it." If you can't fill in the blanks, you don't have a strategy yet — you have a shopping list.
