Delete the Word 'AI'

Take your company’s AI strategy document. Delete every instance of the word “AI.” Read what is left.

If you can still tell which customer you serve, what they need, and how you intend to win, you have a strategy that uses AI. If nothing remains, you have a technology plan in search of a reason.

The underpants gnomes in South Park had a similar problem:

Phase 1: Collect underpants
Phase 2: ?
Phase 3: Profit

Replace the first line with “Deploy AI” and you have a surprising number of corporate strategies.

Start with the business constraint

Strategy requires choices: where to compete, how to win, and what to refuse. “Become AI-first” makes none of those choices. Every competitor can say the same thing.

Consider streaming localization. Human dubbing can cost thousands of dollars per episode-hour and take weeks. That makes it worth doing for flagship titles, while much of the catalog never reaches viewers who need another language.

AI-assisted dubbing can change that calculation. In March 2025, Amazon announced a Prime Video pilot for licensed titles that lacked dubbing support, with localization professionals involved in quality control.[1] The opportunity was to create a dub that would otherwise never exist.

That is a specific bet: reduce localization costs enough to reach audiences the existing economics excluded. It gives a team something to test. “Use AI across content, marketing, and localization” does not.

The test

An AI strategy should answer six questions:

1. What business outcome matters? Better retention in a market, faster implementation, fewer compliance errors, higher conversion. Pick one you can measure.
2. Where does AI help? Name the customer, workflow, and constraint. “Across the business” is too broad.
3. How does it improve your position? Lower cost, greater speed, better accuracy, or access to customers you could not serve before.
4. What has to be true? Identify the data, permissions, evaluation, and failure handling the plan depends on.
5. What will you refuse? Specify the decisions that stay human, the rules that stay deterministic, and the work outside the plan.
6. What evidence would make you stop? Set the success threshold and the conditions for killing the project.

The best strategies remind me of the hypotheses I wrote in college science classes: specific enough to test, and clear enough to be wrong.

The product leader's job

When leadership asks for an AI strategy, start with a business objective that is blocked by a specific constraint. Then show whether AI changes the economics enough to justify the work.

If you cannot explain the opportunity without saying “AI,” you are not ready to fund it.

For a worked example of reducing a proposed automation to a testable investment, see Most Things You Could Build Aren’t Worth Building.

References

[1]: Amazon, “Prime Video Begins an AI Dubbing Pilot Program on Licensed Movies and Series,” March 2025.