Your Company Doesn't Need an AI Strategy. It Needs a Learning Loop It Owns.

There's a statistic from a survey this spring that should be pinned above every leadership team's desk. When WRITER asked 2,400 executives and employees about AI in April, 75% of executives admitted their company's AI strategy is "more for show than for actual internal guidance."

Three out of four. The strategy deck exists. It gets presented. And the people who made it quietly concede it isn't steering anything.

The same survey found the other half of the problem: only 29% of leaders said they'd seen significant ROI from generative AI, and just 23% from AI agents. A lot of strategy, a lot of spend, not much return. Those two numbers are the same story. The strategy is theatre because it was pointed at the wrong thing.

What Nadella Is Actually Measuring

While most companies were writing AI strategy decks, Satya Nadella was quietly reframing what the whole game is about. On Microsoft's autumn earnings call he described the company as "building a planet-scale cloud and AI factory, maximizing tokens per dollar per watt." At Davos in January he went further, floating "tokens per dollar per watt" as a macroeconomic indicator — the idea that future growth depends on producing more intelligence at lower cost.

Strip out the scale and there's a lesson in it for a ten-person business. The input — the raw intelligence, the tokens — is becoming cheap and abundant. When the input is a commodity, your strategy can't be about which model you bought. Everyone can buy the same model. The advantage has to live somewhere the commodity doesn't reach.

Why The ROI Isn't Showing Up

Here's what a "strategy" usually buys: licences. A ChatGPT subscription here, a Copilot rollout there, a pilot with some agent platform. Tools get deployed, a memo goes out, and everyone waits for the productivity to appear. Then the ROI survey comes back at 29% and nobody can quite say why.

It doesn't show up because tools don't compound. The licence you bought this year is the same licence your competitor bought, and it resets to zero every time you cancel it. And leaders keep spending into that gap anyway — BCG found 94% of CEOs plan to keep investing in AI at current or higher levels even without near-term payoffs. More money into the same leaky bucket.

The thing that compounds isn't the tool. It's what your business learns by using it.

The Asset To Own: A Learning Loop

Every time someone in your business does real work with AI — drafts a quote, handles a complaint, scopes a job, answers a client — that interaction contains something valuable: how your business actually does the thing. The wording that wins. The exception that always trips you up. The judgement call only your most experienced person knows how to make.

A learning loop is the discipline of capturing that, on purpose, so the next interaction starts from it instead of from scratch. The prompt that produced a great proposal becomes a template. The way your best estimator reasons about a job becomes written instructions an agent can follow. The context accumulates, and it's yours — it doesn't reset when you switch vendors, and your competitor can't buy it.

That's the asset. Not the subscription. The compounding library of how-your-business-works that sits behind the subscription. A strategy that doesn't build that asset is, almost by definition, just for show.

Why The Labs Want To Run That Loop For You

Here's the tell that this is where the value actually is: the frontier labs are racing to own the loop on your behalf. In May, Anthropic stood up a dedicated enterprise services firm, noting that "companies from community banks to mid-sized manufacturers and regional health systems stand to gain from AI, but lack the in-house resources to build and run frontier deployments." OpenAI built a parallel venture the same week. Both put engineers inside customer offices for the same reason: to build the context layer — the learning loop — that makes the model actually work in that specific business.

They're not doing this because the models are weak. They're doing it because the model was never the scarce part. The scarce part is the accumulated, business-specific context. They'd rather own it than watch you build it yourself — because whoever owns that loop owns the relationship.

What To Do Instead Of Writing A Strategy

If you run a small or mid-sized business, you can't hire a forward deployed engineer. You also don't need one to start the loop. Three moves:

Pick one workflow and make it the system of record. Not "adopt AI." One workflow — quoting, intake, follow-up. Every time you run it with AI, save what worked: the prompt, the inputs, the final version. That folder is the start of your loop.
Write down the judgement, not just the steps. The reason your strategy is for show is that the real expertise lives in someone's head. Get one experienced person to narrate how they actually decide, and capture it as instructions. That's the part the commodity model can't supply and your competitor can't copy.
Own the context, rent the model. Use whatever model is best this quarter — they'll keep leapfrogging each other and getting cheaper. But keep your prompts, templates, and instructions in something you control, so switching models costs you nothing and your accumulated context carries over.

The Honest Read

An AI strategy that names tools and sets adoption targets will keep producing 29% ROI and 75% "for show," because it's optimizing the part that's becoming free. The model is the commodity now. What you've learned about your own business — captured, written down, and compounding — is the only part that isn't.

Stop writing the strategy nobody follows. Start building the loop nobody can buy from you.