Delete the Word 'AI'

Is your AI strategy real, or just another version of the underpants gnomes’ business plan?

Take your company's AI strategy document. Delete every instance of the word "AI." Read what is left.

If a strategy remains, a decision about which customer you serve, how you intend to satisfy their needs, and what makes you different, then you have a strategy that happens to use AI. If nothing remains, you never had a strategy.

South Park nailed this years ago. A tribe of underpants-stealing gnomes had a business plan:

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

Most "AI strategies" are the same plan with the nouns swapped:

Phase 1: Deploy AI
Phase 2: ?
Phase 3: Profit

Most AI Strategies Start in the Wrong Place

Strategy is a set of choices under constraint. Where to play. How to win. What to refuse.

"Deploy AI" is not a choice. "Be AI-first" is not a choice. "AI across every function" is not a choice. Every competitor can say yes to all three, which means none of them decides anything. A real strategy implies what you will not do, and "AI-first" implies nothing you will refuse.

AI capability is not a strategy any more than "cloud" was a strategy in 2010 or "mobile" was a strategy in 2008.

AI strategies fail when they start with technology rather than the business problem. Do not begin with “where can we use AI?” Begin with the thing the business needs to change.

In streaming, one constraint is catalog reach per market. Only a fraction of any catalog is localized into any given language because human dubbing is expensive and slow, costing thousands of dollars and requiring weeks of lead time per episode-hour. Flagship titles get dubbed. The long tail sits unwatched in most of the world.

AI-aided dubbing lowers cost-per-localized-hour, but its strategic value is not "dub everything cheaper." The value is localizing the long tail, which the old economics excluded entirely. Where AI changes the unit economics is specific and narrow.

Amazon was explicit about this when Prime Video launched its AI-aided dubbing pilot in March 2025: the feature runs only on licensed titles "that do not have dubbing support," using a hybrid model in which localization professionals stay in the loop to ensure quality.[1] AI did not replace the dub. It created a dub that would otherwise never exist.

Now write the broken version next to it. "We will be an AI-first studio. AI across content, marketing, and localization." Delete the word "AI," and you are left with "We will be a studio. across content, marketing, and localization." No objective. No constraint. No bet. Nothing to measure, nothing to object to.

The test

Before anyone funds an "AI strategy," run it through these questions:

What are we trying to accomplish? What is the business outcome? More retention in a specific market. Faster implementation. Better compliance coverage. Higher conversion. If the objective is “use AI,” stop. That is not a strategy.
Where will AI actually matter? Name the customer, market, product surface, or workflow where the constraint lives. AI applied everywhere is not a strategy. It is a lack of choices.
How will AI help us win there? What changes if the bet works? Does the work become cheaper, faster, more accurate, easier to scale, more personalized, or possible for a different kind of user to do? If nothing material changes about the economics of the work, the AI is decorative.
What has to be true for this to work? Name the required data, workflows, permissions, evaluation thresholds, and failure handling. Most AI strategies sound plausible because they skip this part.
What are we choosing not to do? Which workflows stay human? Which stays deterministic? Which markets, use cases, and product surfaces are out of scope? A strategy that refuses nothing has chosen nothing.
How will we know whether to keep going? If the bet cannot be measured or killed, it is not a strategic bet.

I still think of the best strategies like the hypotheses I had to write in college science classes: specific, testable, and capable of being proven wrong.

The product leader's job

When leadership says "we need an AI strategy," the worst response is to deliver one. The right response is to translate the request back into the question it was avoiding: which business objective, blocked by which constraint, in which workflow, and where does AI change the economics?