AI strategy on one page
An AI strategy should fit on one page: the business goals AI serves, at most three priority use cases with a metric each, the data and platform choices that support them, the team you need, how you'll govern risk, what ships in the next 3, 6 and 12 months, and what would make you stop a use case. If it needs 40 slides, it isn't a strategy yet.
Why most AI strategies don't work
AI strategies written as technology roadmaps ("adopt agents, build a platform") don't tell anyone what changes for the business. AI is part of your product and operations, not the whole of them, so the strategy should read as business priorities with AI as the method.
Template
Business goals: the 2 or 3 company goals this supports
Use cases (max 3):
1. [process] -> [metric] from [baseline] to [target] by [date], owner [name]
Data: sources needed, owners, access gaps
Platform: build, buy or partner per use case, hosting and privacy rules
Team: roles needed, hire vs upskill vs partner
Governance: risk tiers, review process, owner
Plan: 3 months / 6 months / 12 months deliverables
Kill criteria: what makes us stop a use case
Filling each part
Business goals come from the CEO and the business, not from the AI team. Ask: if two or three numbers improved a lot in the next 12 months, which ones would tell you the company is winning? Abstract answers like "be AI-first" are a sign the goals aren't defined yet.
Use cases come from scoring candidates, and each needs a business owner.
Platform choices follow the use cases, not the other way round. See cloud APIs vs open-source LLMs and build vs buy vs partner.
Kill criteria are what make the strategy honest: for example, no movement in the metric after 90 days in production, or run cost above the value delivered.
Alignment check
Before you publish it, check each use case against four questions: does it fit the company's mission and vision, is there a team strong enough to deliver it, is there a return on it, and is it a real problem someone has today? Anything that fails one of these goes back to the list or gets dropped.
Keeping it alive
Review it quarterly with each use case's metric on the table, and update the page rather than writing a new deck.
From the floor
In the founding session of the AI leadership cohort, I described alignment as putting your mission, your vision, the strength of your team, whether there's a return and whether you're solving a real problem in one place, and making the decision there: this is the right problem statement to move forward with, or we drop it right away.