Test demand before you build the whole business
Use AI to challenge one risky assumption, then test it with real people. Record observations separately from guesses, define what would change your decision, and do not mistake a generated plan for demand.
1. Name the assumption you are testing
“Will this business work?” is too large a question for one experiment. Ask something narrower: do independent consultants spend enough time on proposals to want help? Will a particular buyer pay for an onboarding audit? Can you reach that buyer through a channel you can afford? Write down the assumption and what evidence would make you reconsider it.
2. Ask about the last time it happened
People are better at describing past behavior than predicting future purchases. Ask: “Tell me about the last time you dealt with this.” Follow up with what triggered the problem, what they did, how long it took, and what they paid for help. Avoid asking only whether your idea sounds useful. Let the person describe their experience before showing your solution.
3. Understand the alternative
Your competition may be a spreadsheet, an employee, a freelancer, or simply doing nothing. Ask what the person currently uses and what they like about it. Changing behavior has a cost. Your offer must give them a reason to change that outweighs the effort, uncertainty, and price. Record their words; do not reinterpret every frustration as an intent to buy.
4. Make a small, honest offer
Once you understand the need, describe a pilot you can actually deliver. Include the price, scope, timing, and next step. If something is not built, say so. If you accept payment before delivery, make the delivery date and cancellation terms clear. A proposal or a simple page may be enough. The purpose is to make a real decision possible, not to make an early service look larger than it is.
5. Track the strength of the signal
A like or compliment shows attention. A detailed conversation shows engagement. An introduction to the buyer, a request for a proposal, or a scheduled follow-up may show intent. Payment provides stronger evidence of willingness to buy that particular offer. None of these alone proves repeatable demand. Keep the signals separate so you can see what has actually happened.
6. Decide what to change
If people do not respond, consider whether the audience, channel, message, or timing prevented a fair test. If they talk but do not buy, investigate urgency, scope, trust, price, and the decision process. Do not lower the price automatically. At the end of the test, write a short decision: continue, change one assumption, or stop. Include the evidence and the next experiment. A small test is useful when it changes what you do next.
Use AI to challenge the weakest assumption
Ask AI to find what could make the offer fail, not to tell you whether your idea is brilliant. Start with a specific buyer, a recent problem and the action you hope they will take. Separate what somebody actually did from your explanation of why they did it.
Before choosing a test, score each assumption informally: would the idea fail if this were false, and how little evidence do we have? Start with a high-impact uncertainty. A precise spreadsheet score is not a probability of success. If you have not defined the offer yet, use the offer guide first.
A fictional example: proposal follow-up for independent consultants
Imagine a consultant considering a service that helps small firms follow up after sending proposals. The assumption is not “everyone wants more sales.” It is: “Owners who sent at least five proposals last month lose track of follow-ups, and would consider outside help with that specific task.” This is an invented teaching scenario, not a customer result.
- Known: nothing about demand yet. A tidy service description is not evidence.
- Needed: recent examples of missed follow-up, the current workaround, who owns the decision and what a missed follow-up costs them in effort.
- Alternative explanation: proposals might be losing because the offer is a poor fit, not because follow-up is missing.
- Counterexample: an owner already follows up reliably and still loses proposals. Automation would not resolve that problem.
Give the assistant a bounded review job
Copy this into your preferred assistant after removing identifying information:
“Review this offer as a skeptical planning partner. Audience: [specific buyer]. Offer: [bounded task]. Evidence: [dated, anonymized observations only]. List three assumptions whose failure would undermine the offer. For each, identify the supporting observation, an alternative explanation, the evidence still missing and one inexpensive test. Label any inference as a hypothesis. Do not invent quotes, market sizes, customers or willingness to pay. Ask me for missing evidence before recommending expansion.”
Review its answer against your notes. Reject invented sources and confident interpretations unsupported by the observations. Use the AI output review worksheet to record the corrections. AI can suggest questions; it cannot create genuine buyer behavior by simulating it.
Run a test that could change your mind
For the fictional follow-up service, a first test could be three conversations with owners who recently sent proposals. Ask each to walk through the last proposal, with private details removed. Do not pitch until you understand the process. If none describes a follow-up problem, revise the problem statement instead of building a dashboard. If one asks for help, explore a clearly scoped next step; an interested conversation is still not a sale.
Three conversations are a small learning sample, not a market estimate. Write your decision rule before the conversations so that a friendly response does not become a retroactive “pass.” Record how many suitable people you approached, how many replied and how many completed the conversation. A weak response may reflect the channel rather than the idea.
Your assumption-to-evidence worksheet
- Decision: What will I do differently after this test?
- Assumption: What specifically must be true?
- Observation: What happened, when, and where is the original note?
- Interpretation: What do I think it means? What else could explain it?
- Test: Who is eligible, what will I ask or observe, and what is the time limit?
- Rule: What evidence would support continuing, revising or stopping?
- Result: What actually happened? Leave this blank until the test runs.
- Next move: Record one action, its owner and review date.
Use the free workflow planner to turn the next move into a brief. Keep confidential customer information out of external AI tools unless you have permission and appropriate data controls.
Method and limits
Strategyzer’s Test Card introduction by Alex Osterwalder (March 5, 2015; checked October 25, 2026) recommends defining an assumption, a test, a measure and a threshold before judging an experiment. The worksheet and fictional example above are our own application of that principle, not evidence that this particular service has demand.
If you need a second perspective, the existing free 15-minute fit-call page is an optional enquiry route. Confirm available times there; a call is not a promise of implementation, a paid advisory plan or business results. AI Cofounder and Waymaker share ownership.
Your next move
- Choose one assumption and a one-week time limit.
- Define what would count as a meaningful signal.
- Arrange three relevant conversations.
- Present a clear pilot only when it fits the need.
- Record responses and make one decision from the evidence.
Put the next step into practice.
Use the free starter kit to create a brief, choose a useful task, and begin with your preferred AI tool.
Get the starter kit