AI Can Inform Strategy. It Should Not Decide It.
By Gabbi Stubbs
Intelligence is not judgement. Analysis is not leadership. Optimisation is not strategy.
Strategy is not simply an exercise in analysing information and selecting the optimal course of action. It is an exercise in judgement, choice, conviction and accountability in the face of uncertainty.
AI can materially improve the quality, speed and breadth of strategic thinking. It can expose patterns, challenge assumptions, model scenarios and identify opportunities that humans might otherwise miss.
But it cannot legitimately own the consequences of a strategic decision. The distinction is not that humans are better at processing information — they aren't. It is that strategy requires capabilities and responsibilities that extend well beyond information processing.
Why the limit exists
Your strategy team is almost certainly using AI already. The question for organisations and their boards is no longer whether, but where AI is allowed to influence a decision, who tests its output, and who is accountable when it is wrong. The research gives a clear answer on the principle: AI should support strategy and strategists. It should not decide strategy.
In a 2026 paper in Strategy Science, Russ McBride describes large language models as "mean articulation machines". They detect patterns across vast bodies of text and reproduce the centre of gravity of what has most often been written. That is a real strength for documented, established knowledge. It is a weakness where advantage depends on ideas that are rare, contrarian or not yet written down.
His illustration is the original iPhone. Ask a model trained on 2006 data to design a better phone, he argues, and it would drift towards the consensus of the time: a physical keyboard, a stylus, a removable battery. Apple's design broke several of those norms at once. That bet came from people, not from the average of what had already been said.
What stays human
McBride's findings and the practice of leadership point to six things AI should not own.
Discernment
Fluent output conceals difficulty. Models repeat widely held beliefs, including wrong ones, because frequency outweighs accuracy. A model can say conventional wisdom may be mistaken, but not which belief to reject, or why.
Questioning and original thinking
AI answers the questions it is given. The question nobody has asked, and the idea that contradicts the consensus, are thinly represented in its training data.
Judgement under uncertainty
Probabilities cannot be known, and much of what informs the choice is undocumented: a confidential plan, a relationship, a feel for a market. McBride cites the view that this judgement is conjecture, not calculation.
Purpose, values and trade-offs
Deciding what the organisation stands for, and what it will not do for commercial gain, is a matter of values rather than analysis.
Conviction and political intelligence
A strategy matters only when people believe it and act on it. That means reading a room, building alliances and creating commitment.
Risk and accountability
Someone must set the appetite and own the consequences. A contrarian bet, and accountability when it fails, cannot be delegated to a technology.
One nuance matters. While not all capabilities are exclusively human, increasing AI capabilities do not provide AI with legitimacy, authority or responsibility. Those stay with the people entrusted with leadership and governance.
Where AI should support strategists
AI's contribution should be substantial, not limited to research and administration. Four areas stand out. They draw on what these models do best: search, synthesis, comparison and arithmetic across large bodies of documented information. In each, AI widens what the strategist can see, test and track. It rarely challenges the consensus itself — that remains a human job.
1. Testing the logic of the strategy
Every strategy rests on claims of the form "if we do X, Y will happen". AI can extract those claims from a strategy document, test them against internal and external data for contradictions, gaps and precedent, and surface the conventional objections the plan will meet. It can also spot when a strategy looks reverse-engineered from a fuzzy goal. Two cautions: McBride finds models typically 20–30 per cent behind humans on harder reasoning tasks, and AI cannot see the unspoken politics behind a flag. Treat the output as a prompt for a conversation, not a verdict.
2. Widening the field of view
Strategies are often upended not by a traditional competitor but by a new way for customers to solve their problem. AI can monitor markets, customers, regulation and adjacent technologies at a scale no team can, connect information across disparate sources and flag anomalies and weak signals. The limit is that models favour documented, frequent patterns, so a truly unprecedented threat can still slip through. People decide which signals deserve a response.
3. Calibrating effort and resources
AI can show how effort will land across the organisation, flag teams likely to be oversaturated and call out dates that are impossible for the resources provided. The same arithmetic supports investment prioritisation, capital allocation and workforce planning. It works from what is written down — check the numbers and add what is not, such as real capacity and competing priorities. Leaders decide what to cut, delay or fund.
4. Modelling scenarios and tracking execution
AI can test options against different futures, show which assumptions a plan is most sensitive to and expose unintended consequences. Once a strategy is chosen, it can translate priorities into measurable outcomes, watch leading indicators and flag delivery risks early. People set the assumptions and decide what to do when the signals turn.
Three dangers
Optimising the wrong objective
AI is very good at pursuing a defined goal. It cannot tell you whether the goal is right, and an organisation can become extremely efficient at the wrong strategy.
Reinforcing yesterday's thinking
Models articulate the centre of what has already been written. Without deliberate challenge they reinforce conventional views, and if every competitor asks similar tools, strategies converge.
Outsourcing judgement
The greatest risk is leaders gradually losing the habit and confidence of exercising independent judgement. An AI recommendation is not a substitute for executive or board accountability.
Five questions for the board
The risk is not that AI will make better strategic decisions than people. It is that people will grow so used to accepting AI-generated recommendations that they stop exercising the judgement leadership demands.
The advantage will belong to organisations that use AI to think more expansively and test more rigorously, while keeping judgement and accountability human. The consequential governance question for directors is how to ensure management is using AI to improve judgement rather than outsource it. The boundary will move as the technology develops — revisit it regularly.
Critically, AI should make our strategic thinking better, not make our thinking redundant.
Sources: McBride, R. (2026). Mean Articulation Machines. Strategy Science, 11(1), 31–54. https://doi.org/10.1287/stsc.2025.0439; McBride, R. (2026, 30 August). What Large Language Models Can and Cannot Do in Strategy; Can AI Do Strategy? A Dialogue and Debate, Strategy Science, 2026.
Want to talk strategy?
If this resonates, let's have a direct conversation about your business.