A one-page tool for deciding, before a pilot starts, whether it's there to persuade someone or to prove something.

Taco Bell started testing its drive-thru voice AI in five California stores in early 2024. By July 2026, the AI was taking orders at more than 890 restaurants in 38 states. Each stage of that rollout won approval for the next one. None of those stages produced the answer Taco Bell's chief digital and technology officer, Dane Mathews, gave the Wall Street Journal in 2025, once the AI was in more than 500 drive-thrus. At busy restaurants with long lines, he said, a human might be the better choice.

Taco Bell's pilots won approval at every stage without testing the conditions that later broke them. Everett Rogers described the same split in farm demonstrations decades ago. In Diffusion of Innovations, he separates demonstrations run to evaluate whether an innovation works from demonstrations run to spread it, and notes that often "a single demonstration is expected to fulfill both functions."

The Pilot Planning Canvas turns Rogers' two kinds of demonstration into two kinds of pilot.

Most proof pilots need a budget before they can run, and budgets usually come from persuading someone. Rogers wrote that a demonstration is especially effective when it's run by someone the audience already respects. I've argued before for shipping a visible win, "a battle people can see."

So the canvas treats persuasion as a real job with its own settings. A persuasion pilot gets an owner the audience already trusts, as much visibility as the team can get with the people whose support it needs, and the customers most likely to respond well, labeled as hand-picked so no one mistakes them for typical.

The damage comes when leadership uses a persuasion pilot's result to decide whether to scale. A persuasion pilot runs with the friendliest customers and as much attention as the team can get, so it's set up to go well. Its result shows leadership that the idea can win people over. It can't show them whether the idea will work under everyday conditions, like a busy drive-thru with a long line.

In a survey of 2,002 CIOs and technology executives released in October 2026, Gartner found that only 31% had guidelines for deciding when an AI pilot shouldn't move to production. Those that did were 2.3 times more likely to land in the top-performing group Gartner calls vanguards. Gartner's numbers show an association and can't show that the guidelines caused the better results.

A written rule also needs someone who will enforce it. In a Duke University experiment, eleven managers wrote their own stopping rule before launching a product. None of them killed the product when it failed, and five broke the rule they'd written. So the canvas asks a team to write down, before the pilot starts, the result that would end the project and the person who can make that call. For a proof pilot, it also asks for an owner whose budget and reputation don't depend on a yes.

The Disruption Canvas helps you decide whether an idea is worth pursuing. Strategyzer, the company behind the Business Model Canvas, publishes a Test Card that asks you to write down a hypothesis, the test you'll run, what you'll measure and the result that would prove you right. Neither tool asks whether a pilot is there to persuade anyone.

The Pilot Planning Canvas sits between those two tools. Use it once you've decided an idea is worth testing, and before anyone picks a location, schedules a demo or announces anything. It plans both kinds of pilot, covering who owns it, who it reaches, what conditions it runs under, what result ends the project and what happens if it succeeds.

The canvas is stacked in three layers. Purpose (modules 1 to 3) asks why you're running the pilot, whether it's there to persuade or to prove, and who will act on the result. Conditions (modules 4 to 7) asks who owns the pilot, who the key stakeholders are, who the pilot's audience is, and what advantages the pilot has that production won't. Evidence & Exit (modules 8 and 9) asks what result would end the project and what happens if the pilot succeeds.

Most modules carry two sets of guidelines, one for a persuasion pilot and one for a proof pilot. Follow the guidelines that match your answer to module 2, and write short statements rather than paragraphs.

Treat a hard answer as a signal. If you can't fill in module 1, the decision the pilot was supposed to inform has probably been made already, and you're running a persuasion pilot. If you can't pick one answer in module 2, you're planning two pilots, so fill out a canvas for each. Whichever runs first, leadership should base the decision to scale on the proof pilot's result alone.

A long list in module 7 means the rollout needs more planning than the pilot shows. Write down what the rollout will have in place of each advantage, and you've started that plan.

Before your company's next pilot gets a budget, ask its sponsor to fill out module 2. If they circle prove, ask who picked the locations.