Past Behavior Predicts Future Behavior
My team was interviewing a 30,000-acre corn-and-soybean farmer about how he buys for the operation. Early on he told us, flat out: "I negotiate everything I buy."
So we asked him to tell us about the last major purchase he made for the farm.
Five new trucks. Three of one model, two of another.
Then we asked about the process.
He didn't pause. "I went online to the local dealership, bought them. I already have an account. They brought them out a few days later."
Did you negotiate on price?
"No. Pretty standard. We do trade-ins yearly, handle it all online. I just pay sticker. It's a new truck, not used."
Read that again. The man who negotiates everything he buys just described buying five trucks online, at sticker, without a single back-and-forth.
He wasn't lying. He believed it. But the self-image and the receipts pointed opposite ways. One was who he thought he was. The other was what he'd actually done.
What someone says they'll do is a wish. What they've already done is data.
I call that gap behavioral compatibility. A new solution inherits the odds that people will actually start using it from the behavior it already depends on. If the behavior is already there, you inherit good odds. If it isn't, you inherit nothing.
Here's the version that should change how you scope. If you want someone to buy a thing online, and they have never bought a thing online in their life, your hurdle to success isn't a little higher. It's a different problem entirely. You're not launching a product anymore. You're launching a behavior-change project. Before anyone can use the thing you built, you have to teach them a habit they don't have yet, which is slower, more expensive, and fails far more often than it works.
No existing behavior, no inherited odds.
So before you fall in love with what a customer says they'll do, go find what they've already done. The truck farmer told me he was a hard negotiator. His actual behavior told me he was a fast, frictionless, online buyer who didn't care about a few hundred dollars on a truck. If I'm selling him anything, I build for the second guy. The first guy doesn't exist.
This isn't a hunch. Three fields landed on it independently.
You don't have to take my word that the past beats the wish. Three separate fields landed on the same finding from three different doors.
Psychology. Judith Ouellette and Wendy Wood ran a 1998 meta-analysis in Psychological Bulletin, titled, almost too on-the-nose, "The Multiple Processes by Which Past Behavior Predicts Future Behavior." Their finding was blunt. Past behavior predicts future behavior, hardest of all when the behavior is frequent, habitual, and the situation looks like last time. The truck-buying was yearly and routine. That's exactly the kind of behavior that repeats. His stated self-image wasn't.
Economics. Paul Samuelson called it revealed preference back in 1938. Stated preference is what people tell you in a survey or an interview. Revealed preference is what their actual money, time, and clicks already show. When the two disagree, the money wins. Researchers even have a name for the distance between them. The say-do gap. Designing a product around what people say while ignoring what they do is one of the oldest and most expensive mistakes in the book.
Adoption. Everett Rogers got there first, and he named it compatibility. In Diffusion of Innovations, compatibility is how well a new thing fits the existing habits, values, and past experience of the people you want to start using it. Behavioral compatibility, from up top, is the same bet aimed at a single product. Rogers found that compatibility explains a large share of why some products spread fast and others die on the shelf. That "different problem entirely" is just Rogers in plain English. An app for people who don't use apps is a low-compatibility bet, and low-compatibility bets are the ones that die on the shelf.
Three fields. One conclusion. Watch what people do, not what they say.
The fix is in the question you ask.
Most of this damage gets done in the interview, with a question that sounds smart and isn't. "Do you negotiate?" "Would you use an app that does this?" "How much would you pay?" Every one of those asks the customer to predict their own future, and people are wildly optimistic about who they'll be tomorrow. You'll get a polite, confident, useless answer. Rob Fitzpatrick wrote a whole book on dodging exactly this, The Mom Test. The core move: ask about specific things they've already done, not what they think they might do.
Watch what that does to the truck conversation. If I'd asked "do you negotiate?" I'd have walked away with "yes, always," and built the wrong thing. Instead I asked: "Tell me about the last major purchase you made." That one question can't be answered with a self-image. It can only be answered with a receipt.
So swap the questions:
- Instead of "would you use an app that does this?" ask "walk me through the last time you had this problem. What did you actually do?"
- Instead of "how much would you pay?" ask "what does this problem cost you today?"
- Instead of "do you do X?" ask "tell me about the last time you did X."
Same instinct every time: drag the answer out of the past, where there's a real record, instead of the future, where there's only a wish.
Surfacing real past behavior instead of polite predictions is a skill, and it's mostly in how you run the conversation. So before you build anything, get good at asking the question that gets you the receipt instead of the wish. That's what Talking to Customers is for.