Playing to Win When Everyone Has AI

The easiest mistake to make with AI is to confuse everything you can now do with a reason to do it.

AI can help you build more products, produce more content and automate work that once required a larger team. Opportunities that seemed out of reach suddenly look plausible. Choosing among them becomes a strategy problem of its own.

Roger Martin and A.G. Lafley’s Playing to Win provides a useful discipline: make five connected choices about what winning means, where you will compete, how you will win, which capabilities you need and which management systems will support them. The strength of the framework comes from how those choices fit together. Your chosen customers should have a reason to prefer you, and your business should be organized to deliver on that reason. Martin’s explanation of the framework^[1]

Applied to AI, the questions remain useful. What deserves reconsideration is the answer to each.

More possibilities make choosing where to play more important.

If AI makes it feasible for a small team to serve ten markets, that team still has limited attention, relationships and resources. It needs to decide which customers it understands, which problems are worth solving and where it has a credible route to those customers.

The ability to build something is one condition for entering a market. Customers caring enough to buy it is another.

Access to AI puts more pressure on the explanation for how you win.

When competitors can use similar models, access alone gives customers little reason to choose one company over another. Advantage has to emerge from the broader offering: how the technology works with expertise, customer relationships, distinctive information or an unusually good understanding of a particular problem.

Consider a hypothetical small research firm. AI might help it produce reports faster. If its competitors can do the same, clients may soon expect faster delivery from everyone.

The firm could instead make a specific strategic choice: serve regional manufacturers deciding which export market to enter. It could combine AI-assisted research with local relationships, verified commercial information and judgment about what would change the client’s decision.

That choice gives the technology a purpose. It also guides what the firm declines. A project can be technically easy to complete and still pull the business away from the capabilities it needs to build.

Capabilities and management systems have to support the promise.

For that research firm, producing a plausible report is only one part of the work. Delivering useful advice requires recognizing weak evidence, investigating contradictions and understanding the client’s constraints.

Its management systems should reinforce those capabilities: clear responsibility for recommendations, source-checking routines, feedback from clients and measures of whether engagements are profitable. Faster production has value only if the complete service remains useful and economically viable.

This is where the framework becomes particularly practical. A compelling AI demonstration can create confidence before a business has worked out how to deliver consistently. The final two choices force that work into the strategy.

What changes most is how you test the assumptions.

Our view is that AI gives companies a reason to examine the expected life of their advantages more carefully.

If your advantage depends on doing something cheaply, what happens when competitors gain the same cost reduction? If it depends on superior model performance, what happens when another model catches up? If customers can do part of the work themselves, which part will they still pay you to handle?

These questions should influence investment and positioning early. They also provide a basis for deciding when to revisit the strategy. A new model release matters when it changes the economics, customer behavior or capabilities on which your choices depend.

Playing to Win already asks leaders to make choices under uncertainty. Applying it to AI means being explicit about what must remain true for those choices to work, and gathering evidence while there is still time to adjust. Martin on testing strategic assumptions^[2]

A useful place to start is with one question: If your competitors had the same AI tomorrow, why would your chosen customers still choose you?