What game theory reveals about how business leaders make decisions without all the facts

For many years, I believed that solid decision-making depended on solid information. The more research you do, the more likely you'll be able to identify the correct path. So whether it's choosing a candidate to hire, setting the price of your products, or entering a new market, I always tried to create as complete a picture as possible. After spending some time looking at how AI systems develop decision-making algorithms within competitive environments, I now realize that my approach was completely backward.

AI decision-making systems are among the most effective systems in existence. Yet they don't attempt to predict the future. They don't determine "the" correct decision. Rather, they seek to minimize regret — or specifically, the amount of better choices (across billions of simulations) they may have made with each and every alternative decision. Through that process, these systems converge upon strategies that can never be leveraged against them.

Business leaders can also learn a great deal from this decision-making style.

The expected value model

Most business decisions rely on outcomes. Did you land that big client? Great decision. Was that product a dud? Bad decision. This type of thinking, however, has been proven to be fundamentally defective for many years. In highly competitive markets with limited visibility into your competitors' knowledge and plans, you can make a decision that is technically flawless and still lose. On the flip side, you can make a lousy decision and still win. Since the end result is dependent on factors that were unknown to you at the time you made your decision, there's really no basis for using the outcome as proof of the validity of your process.

A colleague called me recently lamenting his dilemma regarding a job opportunity. Better title, 30% salary increase, new city, no clue what his new team was like. He'd created a weighted spreadsheet; yet he still hadn't decided.

I said he had a decision-theory problem, not a spreadsheet problem. The appropriate framework isn't "what will happen?" It's "If I had this same class of decision 1000 times, would I be better off saying yes or no?" You're not optimizing for one particular outcome. You're optimizing for the process.

Regret minimization in practice

The AI method that finally cracked the puzzle of competitive, imperfect information-based decision-making is called Counterfactual Regret Minimization (CFR). CFR is elegantly simple: after every decision, ask yourself "How much regret would I have if I had chosen differently?" Then aggregate those regrets across millions of iterations. The strategy that emerges cannot be exploited by anyone else.

I find this concept extremely relevant to business. We all feel regret — "I wish I had taken that deal," "I wish I hadn't hired that guy." An AI system simply thinks through its regrets more systematically and without the emotional baggage.

Pricing strategy: Your pricing becomes independent of the action(s) taken by individual competitors at equilibrium. Your pricing works fairly well under any competitive reaction. Most companies act reactively — "They reduced their price; therefore we should too!" That's not a sound strategy for pricing. A strategy that is game-theoretically consistent takes into account all potential reactions from competitors prior to taking action.

Negotiation: Good negotiators vary their negotiating approach (sometimes aggressive, sometimes cooperative) at pre-determined rates. If you only negotiate aggressively when you have bargaining power, opponents will recognize your aggressive postures and concede immediately anytime you use them. As such, you become predictable. The game theory solution is to randomly intermingle different approaches so that your opponent can't infer your true position from your actions.

Investment decision-making: Compare the likelihood of success with the relative magnitude of the reward compared to the cost of an investment. A new product line costs $200K and generates $1.5 million in revenue if successful. At what probability of success is this investment worth making? Approximately 13%. When viewed through this lens — "Do I think there is at least a 13% chance?" — most investments appear to be relatively straightforward decisions.

Marginal edges compound

While top-of-the-line AI systems may not win every competition, nor win most competitions — they win slightly more often than they lose, consistently over tens of thousands of decisions. The margin may only be 3-5%, which may seem negligible on any given individual outcome; however, that 3-5% compounded annually results in an enormous difference over the course of a year.

I began applying this to almost every aspect of my life. I did not need to dominate every negotiation. I needed to be slightly better prepared than everyone else involved in negotiations — consistently. I didn't need every networking event to produce a sale — I merely needed to build the habit of attending events. The cumulative impact of small margins is grossly underestimated in business.

Advantage due to information asymmetry

Studies conducted on how AI identifies and extracts strategic information in competitive settings have resulted in methodologies that apply directly to identifying strategic information about competitors. Modern AI creates a probabilistic model of all opponent strategies at any point in time and updates it dynamically based on actions taken by opponents.

This translates to business: anytime an opponent launches a new product, reduces prices or lists a job opening — you should update your model of their strategy based on what you've seen. A competitor placing 3 senior-level postings for machine learning engineers raises the probability they're working on developing an artificial intelligence capability. Dropping enterprise pricing levels increases the probability their sales pipeline is weak. With each piece of evidence received, you reduce the range.

I'm aware of a SaaS company founder who establishes a "strategy range" for each competitor's moves into a quarterly report — borrowing language from game theory directly. It's not paranoia — it's structurally rational competitive thought processes.

Embracing variance

My previous business partner couldn't cope with poor performance during a quarter. One quarter less than predicted and the entire strategy was open for review again. Emergency meetings, restructuring and/or pivoting.

Game theory teaches us that variance exists — and can't be avoided — even when executing optimally. Therefore, it is inevitable that you'll experience losing streaks regardless of how well you execute your plan. The question isn't "Did we have a terrible quarter?" — it's "Is our strategy sound over 100 quarters?" If yes, then the subpar quarter was nothing more than random variation. If no — then you should have recognized it long before the subpar quarter occurred.

All the business executives I have worked with exhibit this quality. None panic at short-term results since they know their methods work over substantial sample sizes. That's game-theoretic thinking — and it may be the single most important framework any executive could ever embrace.

The uncomfortable comfort

It's impossible to eliminate uncertainty entirely; uncertainty is merely a condition under which all business decisions are made. Every market entry, every employee hire, every pricing decision and every partnership involves incomplete information and educated guesses.

Those who successfully manage uncertainty aren't smarter or luckier than others; rather they continue to make good decisions under conditions of uncertainty repeatedly until they no longer fear uncertainty itself. This is what decades of game theory research formally define — and is applied much further than competitive situations where it originated.

The author writes about AI, game theory, and their applications in business strategy.