AI won't automatically improve your strategic decisions. A few conditions decide whether it does.

You gave AI a seat at the strategy table expecting sharper decisions, and some of the time that is exactly what you got. But not always. Sometimes the same tool handed you a fluent, well-argued recommendation that turned out to be confidently wrong, and the unsettling part was that you could not tell the good call from the bad one in advance. The tool did not change between those two moments. The conditions around it did, and that is the whole story.

Here is what the evidence says. A 2025 systematic review in the journal Administrative Sciences examined how generative AI affects strategic decision-making in entrepreneurial ventures, and its central finding is a conditional one. AI can improve strategic decisions, but whether it does depends on a set of factors: external ones, above all the technological and market environment you operate in, and internal ones, your data, your capabilities and your culture. AI is not a decision upgrade you can simply buy. It is an amplifier whose sign, better or worse, is set by the conditions you put around it.

Why doesn't AI automatically improve a decision?

Because a decision is only as good as what informs it, and AI informs it from your data and your context. Point a capable model at thin, biased or out-of-date information and it will produce a recommendation that is beautifully written and quietly wrong. The fluency is the trap: an AI answer arrives sounding more certain than a human's hedged judgement, so it is easy to mistake confidence for grounding. In a fast, uncertain market, an AI prediction can also age quickly, so a call that was right last quarter is stale this one.

This is the same pattern behind why most organisations fail at AI adoption: the tool works, and the conditions to use it well are assumed rather than built. And it shows up in the numbers. KPMG's 2026 finance research found that organisations able to produce audit evidence for how an AI output was reached report significant error reduction at 33%, against just 6% for those that cannot. Same class of tools; more than five times the improvement, decided entirely by the conditions around the decision.

What conditions actually make AI-assisted decisions better?

Two sets, external and internal, and you control more of them than it feels like. Externally, the quality of your data environment and the volatility of your market decide how much an AI prediction is worth. Internally, the discipline to question an output, the governance to trace it, and a culture that treats AI as an adviser rather than an oracle. The internal ones are where most organisations are thin: Deloitte found only 21% have a mature model for governing autonomous AI, and Stanford's 2025 AI Index found that while organisations widely recognise AI's risks, many do not actually act on them. Recognition without the conditions to manage it is where an unexamined AI decision does its damage.

How do I build those conditions?

By treating AI as the last thing you add to a decision process, not the first, and preparing the ground it lands on.

1. Fix the inputs before you trust the tool. AI inherits the quality of your data and context, so a decision built on poor inputs will be confidently wrong. Sort the inputs first.
2. Match AI to the decision type. Data-rich, repeating decisions gain most; novel, high-judgement, first-of-their-kind calls still need a human to lead and AI to assist.
3. Design the loop, not just the prompt. Decide how AI feeds the decision, where it must stop for a human, and how you would catch it drifting, in the spirit of designing the loop the work runs in.
4. Build a team that can argue with it. A group that can challenge an AI recommendation makes better decisions than one that defers to it. Reward the challenge.
5. Keep a human accountable for the call. AI informs; a named person decides and owns the outcome. These are exactly the oversight questions a board should be asking.

What does this change for me as a leader this quarter?

It changes what you buy and in what order. The temptation is to acquire the AI and expect better decisions to follow. The evidence says reverse it: build the conditions, the clean inputs, the governance, the questioning culture, and the same AI becomes a genuine decision advantage. Skip them, and you have bought a very persuasive way to be wrong faster.

That is the quieter discipline behind the end of business as usual: the tools are now universal, so the advantage sits with the organisations that prepare the ground for them. Put the conditions in place, keep a human owning the call, and AI stops being a gamble on your strategy and starts being a genuine sharpening of it.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.