The difference between AI strategy and AI deployment.

Ask most mid-market firms about their AI strategy and you will get a description of their AI deployment: the tools they have bought, the functions they have automated, the pilots underway. These are deployment decisions. They are not a strategy, and the gap between the two is where most AI programmes quietly fail.

Strategy is the prior question. It asks what kind of company the firm intends to be once routine cognition is absorbed into its operating model — which judgements remain human by design, which run under review, and which can be automated outright without anyone losing the thread of why. Deployment is the answer to a strategy. Run in the other order, the firm ends up, two years in, with an operating model no one designed and no one can defend to a board.

The decision that matters least is which model to use; it changes monthly and the differences are narrowing. The decisions that matter most are organisational and they are durable: where accountability sits when a system, not a person, was the proximate cause; which decisions a serious firm should still want a human to make even when a machine could make them faster; and how the firm explains all of this to a board that has to sign off on it.

The board conversation, then, is not a technical one, and treating it as technical is itself a failure of strategy. The board does not need to understand the model. It needs to understand the operating model — the map of decisions by who or what makes them, and the review gates around the consequential ones. That is a conversation a board can actually have, and the firms that have it early are the ones that deploy without regret.