Say vs Do: Why Actions Beat Words in Validation

What people say they will do is weak evidence; what they actually do is strong. Stated intentions, opinions, and predictions about future behavior rarely come true, while real actions — especially costly ones — reliably forecast what comes next.

Quick Answer: "Say" evidence is what people tell you: opinions, compliments, and predictions about how they will behave. "Do" evidence is what people actually did — a purchase, a signup, a workaround they built themselves. Say evidence is weak because stated intent is a poor predictor of behavior; do evidence is strong because past action forecasts future action. Design tests around doing, not talking.

Every founder has heard "I would definitely use that." Almost none of the people who said it ever showed up to buy. The gap between what customers say and what they do is the most expensive misread in early validation — and it is avoidable once you know which kind of evidence you hold. This is a core principle of Testing Business Ideas by David Bland and Alex Osterwalder: evidence from what people do is stronger than evidence from what they say.

Say vs Do: The Evidence Strength Comparison

Say and do are not two shades of the same evidence — they sit at opposite ends of a strength scale. One captures a free-to-give opinion; the other captures a decision that cost something. The distinction determines whether feedback should change your plans or be politely ignored.

The table below contrasts the two on the dimensions that decide how much weight a piece of feedback deserves. Use it to grade any input before you act on it.

Takeaway: When you are unsure whether feedback is worth acting on, ask one thing — did the person do something, or just tell me something? Words alone belong near the bottom of the scale, however warm they sounded.

Why Interviews Produce Say Evidence

Most customer interviews generate say evidence by default, because you are asking people to imagine the future instead of watching them act. Questions like "Would you use this?" or "How much would you pay?" invite a prediction, and humans are famously bad at predicting their own behavior. The answer feels like data. It is a guess wearing a suit.

Rob Fitzpatrick's The Mom Test diagnoses the failure precisely: the moment you ask someone's opinion about your idea, you invite them to be nice to you. Compliments are worthless as evidence. So are hypotheticals ("Would you...?") and generic claims ("I always...") — they pull answers toward what sounds good rather than what is true.

The fix is to interrogate the past, not the future. Fitzpatrick's rule is to ask about specific things that already happened:

• Ask what they did, not what they would do. "Walk me through the last time you faced this problem" beats "Would you use a tool that solved this?"
• Ask for facts, not opinions. "What are you using now, and what does it cost you?" beats "Do you think this is a good idea?"
• Ignore compliments entirely. "I love it" is a red flag, not a green light — steer back to their actual behavior and spending.

Even done perfectly, an interview about the past is still someone reporting on their actions rather than taking one in front of you. That makes it stronger than a hypothetical, but it is not yet true do evidence. Learning to separate the signal from the noise in validation starts with recognizing that most interview output is a report, not a measurement.

How to Turn a Say Test Into a Do Test

You convert say into do by adding a real ask — a moment where the person has to spend something to continue. The redesign is usually small. Instead of ending a conversation with a question, you end it with an opportunity to act, then measure who acts.

Replace the survey question with a landing page. A survey asks "would you sign up?" A landing page with a real signup form measures who actually does. The click that costs an email address is worth more than a hundred verbal yeses.

Replace "would you pay?" with a pre-order. Testing Business Ideas frames this as choosing stronger evidence on purpose — a call-to-action experiment over an interview when the stakes are high. Money on the table before the product exists is the classic do signal. It carries skin in the game, the cost that makes an action predictive.

Replace "is this useful?" with a concierge test. Deliver the outcome manually to a handful of people and see whether they keep coming back and paying. Repeat behavior is do evidence at its strongest.

Recruit the right people first. A costly action from the wrong audience is still weak. If you need a foundation, the complete guide to customer research for founders covers how to find and talk to the people whose actions actually matter. The point of every redesign is the same: engineer a moment where saying yes has a price.

When Say Evidence Is Still Useful

Say evidence is weak for prediction but genuinely useful for direction. Dismissing it entirely is a mistake in the other direction — words are how you discover which experiments are worth building at all. The rule is to use say evidence to form hypotheses, never to confirm them.

Use interviews to find the problem. What people describe about their past struggles, workarounds, and spending tells you where a real problem might live. That is discovery, and The Mom Test is built for exactly this — mining specific stories to locate pain worth solving.

Use say evidence to shape the do test. The language customers use, the alternatives they mention, and the moments they describe give you the raw material to design an experiment they will respond to. Say evidence points the flashlight; do evidence is what you actually see when it lands.

Never let say evidence be the final word. The failure mode is treating a warm interview as proof and skipping the costly test. Opinions can green-light building the experiment. Only actions can green-light building the product. Tools like Edmired can help you track which evidence is say versus do, but demanding a real action before you commit is a discipline you keep yourself.

Key Takeaways

• What people do beats what they say. Testing Business Ideas ranks real actions above stated intentions, because past behavior predicts future behavior and talk does not.
• Interviews default to say evidence. Asking "would you use this?" invites a prediction, and people are poor at predicting their own behavior.
• Compliments are worthless. The Mom Test warns that any opinion about your idea invites politeness — steer every conversation back to specific past facts and spending.
• Ask about the past, not the future. "Walk me through the last time..." produces far stronger evidence than "Would you...?"
• Turn say into do by adding a cost. Replace survey questions with landing pages, "would you pay?" with pre-orders, and "is this useful?" with concierge tests.
• Use say evidence for direction, not confirmation. Words are excellent for discovering problems and designing experiments — but only actions should green-light the build.

Frequently Asked Questions

Why is what people say weaker evidence than what they do?

Because stated intent is a poor predictor of actual behavior. When someone says "I'd buy that," they are guessing about a hypothetical future while feeling social pressure to be encouraging. Testing Business Ideas ranks evidence from what people do above what they say precisely because an action already happened and cost something — a purchase or signup forecasts future behavior in a way a prediction never can.

Does this mean customer interviews are useless?

No — interviews are valuable for discovery, not confirmation. The Mom Test shows how to run them well: ask about specific past behavior and facts rather than opinions about your idea, and ignore compliments entirely. Good interviews reveal where a real problem lives and how customers describe it, which tells you what do-based experiment to build next. They simply cannot prove demand on their own.

How do I turn stated interest into a real do test?

Add a moment that costs the person something. Instead of asking "would you sign up?", put up a landing page and measure real signups. Instead of "would you pay?", offer a pre-order and see who pays. Instead of "is this useful?", deliver the outcome manually and watch who returns. Each redesign replaces a free opinion with a costly action you can observe.

Is a strong verbal commitment ever enough to build on?

Rarely, and never as the final word. Even an enthusiastic "I will definitely buy this" is still a say — a prediction made when saying yes is free. Treat it as a reason to run a costly test, not as validation itself. If the person will not take a small real action now, such as a deposit or a booked call, their words are unlikely to become behavior later.