When someone counts their change at the bakery, they are using mathematics. Nobody would put it that way, and for that particular task, a reasonable intuition for numbers does the job perfectly well.

Something similar happens when you ask for a pay rise, decide how much to reveal in a negotiation, or try to persuade two colleagues to commit to the same project. You think about what other people want, how they might respond, and what that means for your next move.

These are the situations game theory studies: decisions whose consequences depend on other people’s decisions, including their expectations about yours.

You do not need to have studied the subject to reason about them. Experienced negotiators, founders and managers make good strategic calls every day without a single payoff matrix. The discipline has no monopoly on strategic thinking, any more than mathematicians have a monopoly on numbers.

But the analogy works in both directions. Being comfortable with numbers does not qualify you to price an insurance portfolio. Being comfortable around a negotiating table does not mean you have worked through the incentives of everyone sitting at it.

My argument is that we rely on that distinction far too little, and that AI gives us a reason to take it more seriously.

When intuition becomes expensive

Splitting a restaurant bill needs no actuary. Pricing insurance against rare, correlated losses requires rather more than a feel for what usually happens. Years of apparently successful decisions can conceal a risk that has simply not materialised yet.

Strategic problems have their own sources of difficulty. Several people can move more than once. Important information stays private. A commitment changes what someone else can demand later. You may get only one attempt, with a large difference between an acceptable outcome and a good one.

A familiar conversation with modest stakes rarely needs a formal model. A market entry that could provoke retaliation, a partnership involving irreversible investment, or a change to incentives across an organisation deserves more scrutiny.

Consider a partnership that looks attractive to both sides. One company must invest first; the other can renegotiate once that investment is sunk. A spreadsheet can show substantial value creation while leaving unanswered whether either side has a reason to honour the arrangement.

Or consider a bonus scheme that achieves exactly what it rewards while making the business worse. The employees may understand the incentives perfectly. The person who designed them may be the one who has missed the strategic problem.

In both cases, the expensive assumption is about what other people will do once you have made your decision.

What formal training adds

A simple guessing game makes the issue unusually clear. In the two-thirds game studied by Rosemarie Nagel and others, participants choose a number between 0 and 100. The winner is whoever comes closest to two thirds of the average.

If you expect an average of 50, you aim around 33. If you expect others to reason that way, you might aim around 22. Continue the reasoning and you reach the equilibrium where everyone chooses zero: nobody can improve their result by changing their choice alone.

The practical analysis also asks what this particular group is likely to choose. In later newspaper experiments, 81% of the 422 participants whose explanations identified the equilibrium nevertheless chose a higher number. They allowed for other people reasoning differently. [1]

This is precisely the distinction applied game theory equips you to work with. You can identify an equilibrium, examine the assumptions that support it, and work out how your decision should change when the people involved have different beliefs or reason differently.

In business, those differences are everywhere. Your competitor may value market share more than this year’s margin. Your negotiating partner may face an internal deadline you cannot see. Someone rejecting an apparently attractive offer may have an alternative you have underestimated.

Formal training gives you a disciplined way to investigate these possibilities. You specify the relevant choices, information and incentives, then examine which assumptions actually change the recommendation. Commercial experience supplies essential context; the analysis makes it possible to challenge the reasoning before acting on it.

The Harvey Specter problem

In Suits, Harvey reads incentives, anticipates the next move and works out what the other side cannot afford. He makes it look like a particularly well-dressed form of instinct.

He also has an advantage unavailable to the rest of us: the people writing his brilliant tactics write his opponent’s response.

Real negotiators have to live with a less accommodating arrangement. Yet “he’s a natural” still passes for an explanation of why someone’s advice should be trusted, and twenty years of experience can end a discussion that ought to be beginning.

Experience deserves respect. Its reliability also depends on what someone has had the opportunity to learn. Daniel Kahneman and Gary Klein identified predictable patterns and opportunities for feedback as important conditions for reliable intuition. Confidence itself does not establish either. [2]

Closing a deal tells you that the deal was possible. It tells you much less about whether a different approach would have produced a better one. In a decision you make only once, separating good judgement from good fortune is particularly difficult.

That is a reason to examine the reasoning behind a recommendation. A confident answer becomes more useful when the person giving it can explain what would have to be true for it to work, and what they would recommend if it were not.

Sometimes you can improve the rules

The contribution can go further than anticipating a response. Often, you can change what people have an incentive to do.

A partnership can link investments to milestones and agree how later decisions will be made. A company can redesign a bonus scheme so that helping a colleague no longer damages an employee’s own prospects. A negotiation process can make it easier to reveal useful information without giving away everything at once.

This is why I find the familiar picture of a game theorist as someone calculating how to beat an opponent so restrictive. A substantial part of the work concerns making cooperation possible and keeping it worthwhile as circumstances change.

The result may be a recommendation, but it may also be a contract, an allocation rule, a process or a tool. The value lies in what people have reason to do once it is in place.

AI expands what we can examine

For practical analysis, there has always been a limit to how much can be investigated before a decision has to be made. A team selects the most important interactions, works through a manageable set of assumptions and uses judgement for the rest.

AI gives trained analysts additional capacity to explore that problem. It can assist with implementing models, generating alternative cases and revising an analysis as new information arrives. Those outputs still need checking, but the opportunity is substantial: more of the strategic situation can receive serious attention within the decision window.

For a market entry, that could mean examining several credible competitor responses rather than carrying one central assumption through the entire forecast. For a partnership, it could mean testing whether the arrangement still works when demand disappoints, one party gains bargaining power, or the timing of investment changes.

The improvement worth pursuing is a recommendation that has survived more demanding scrutiny.

There is already a concrete example of combining AI with structured strategic reasoning. Meta’s CICERO system brought together a language model, a planning engine and models of other players’ likely behaviour to play Diplomacy. Its architecture connected what the system said to what it expected others to do. A board game is much more tightly specified than a business decision, but the example shows why language capability and strategic analysis belong together. [3]

The important work still includes deciding which situation to model. Who can actually commit? Which information is private? Does a threat remain credible when the time comes to carry it out? Solving and validating a complex model can itself be difficult; answering the wrong question more fluently does not help.

As analytical tools improve, I expect more decisions to justify this depth of strategic work. The opportunity is to apply expertise more widely and get more from it.

Why this matters for advisory work

This has a particular implication for consulting, though it extends well beyond it.

As AI takes on more research, coding and document production, firms have to be clearer about the judgement embedded in their advice. An elegant market forecast remains incomplete if it assumes competitors will politely leave the opportunity alone. A detailed implementation plan can fail because the people expected to carry it out have reasons to resist it.

People trained in strategic interaction can help identify those problems while the decision is still open to change. Working alongside sector specialists and the people responsible for implementation, they can examine reactions, test commitments and design arrangements that work with the incentives involved.

I think we need far more of that expertise across business and public policy. Its relevance extends from pricing and partnerships to organisational design and the rules governing automated decisions. We have spent too long treating it as an academic specialism with a few commercial applications.

The case for wider use should ultimately be judged by the decisions it improves. Does the analysis uncover a dependency the team had missed? Change a commitment before it becomes expensive? Produce an arrangement that still works when circumstances turn against it?

Those are useful tests of the work, and increasingly capable tools give us more opportunity to do it well.

We all use mathematics. Nobody takes that as a reason to leave difficult quantitative decisions to a good head for numbers. We all think strategically, too. The consequences deserve the same respect.

Sources

[1] Antoni Bosch-Domènech, José G. Montalvo, Rosemarie Nagel and Albert Satorra (2002). One, Two, (Three), Infinity, …: Newspaper and Lab Beauty-Contest Experiments. American Economic Review, 92(5), 1687–1701. Includes discussion of Nagel’s original laboratory experiments and the newspaper participants’ explanations.

[2] Daniel Kahneman and Gary Klein (2010). Strategic decisions: When can you trust your gut?. McKinsey Quarterly. Interview discussing their research on the conditions for intuitive expertise.

[3] Meta Fundamental AI Research Diplomacy Team et al. (2022). Human-level play in the game of Diplomacy by combining language models with strategic reasoning. Science, 378, 1067–1074. See also the official system description.