We have read a lot of AI strategies. Most are correct and none of them ship. The gap is not intelligence. It is distance. The strategy is written by people who will not build it, for people who were not in the room, about systems nobody in either group has opened.
The handoff is the failure
A strategy that ends in a handoff has already decided what happens next: a second team rediscovers the constraints the first team never saw. The legacy interface with no documentation. The regulator who needs the audit trail in a particular form. The controller who will not approve anything she cannot read.
We do not hand off. The strategists who decide what to build and the engineers who build it are the same team, embedded in the same organization, from the first workshop to the production release.
What that changes about the strategy itself
It is written in the client's constraints. Your systems, your data, your regulator, your calendar. A roadmap that ignores any of them is a wish.
It sequences by readiness as much as value. The highest-value use case is often the one the systems cannot carry yet. We rank on both axes and start where the load can be borne.
It names the approver. Every workflow in the plan has a person in the business who will sign off on the agent's work. If that name is blank, the item is not ready.
It ends in production, not a deck. The deliverable of a strategy engagement is the first workflow running under governance, and a plan for the next five that inherit it.
The operating model is the strategy
AI does not change a business by being added to it. It changes a business by changing where judgment sits, who approves what, and what a team is for. That is an operating-model question, and it belongs in the strategy on day one, not in a change management annex at the end.
Strategy that is written for the people who deliver it gets delivered. That is the whole method.