There is no cheaper sentence in business than “I would buy that.” It costs the speaker nothing, it pleases whoever asked, and it brings a slightly awkward question to a close. It is also, as a rule, perfectly sincere, which is what makes it so expensive for everybody else.
The bill arrives later, and in a familiar form. A survey goes to 400 customers, asking whether they would pay $30 a month for an analytics add-on. Thirty per cent say yes. That is 120 customers and some $43,000 a year, and on the strength of it a team spends a quarter building the thing. It launches. Forty-one people buy it.
Nobody lied. Ask any of the 120 and they would tell you they meant it at the time. Researchers call the difference the say-do gap, and they have been measuring it since long before anyone sold software. The measurements are the useful part, because a gap of known size can be planned for.
Half, or a third
The cleanest figures come from American households that were asked about large purchases and then followed for a year to see what they did. About 10% said they would buy a car within twelve months. About 5% bought one. For personal computers, at that time a new sort of product, 15% said yes and fewer than 6% went through with it. For the familiar thing, roughly half the promises became purchases. For the novel one it was nearer a third.
Prices swell in the same way. Across 77 studies of consumer goods, people named prices about 21% higher, on average, than they paid when real money was involved, and the gap was wider for expensive and specialty products.
It would be easy to conclude that people are careless with the truth. The research is more forgiving than that. To begin with, a hypothetical yes is free. In one experiment with a cleaning product, only 5% of people picked “none of these” when the choice was hypothetical. When they knew they might really have to buy what they chose, 19% did. The product was the same. Only the price of the answer had gone up.
Often, too, there is no plan to report. Asked whether they would buy your add-on, people frequently have no stored intention to look up. They assemble one on the spot, out of the mood of the moment and the way the question was put.
And the largest part is the most human. Most of the gap comes from people who fully mean to act and then don’t. They underestimate the effort, or the moment passes, or something else wants the money. In a company the budget is cut, or the spreadsheet the team already uses turns out to be good enough for another year.
The traffic runs the other way as well, which is noticed less often. “No” is unreliable too. In many studies, most purchases came from people who had reported no plans to buy. People without plans rarely buy, but there are so many more of them that they end up supplying most of the customers.
The discount
None of this makes the question worthless. In data on cars and wireless services, forecasts that used what people said they intended beat forecasts that merely extended past sales. The answer carries information. It just cannot be taken at face value, any more than a builder’s estimate can.
So discount it. The household studies suggest by how much.
Return to the add-on. Four hundred customers were asked and 30% said yes, which is 120 people. The product was new to them, so the honest expectation lay between a third and a half of that: 40 to 60 buyers. The 41 who turned up were not a disappointment. They were the forecast, had anybody done the sum.
The price deserves the same treatment.
Thirty dollars becomes about $25, and you should cut further for anything expensive or unusual.
These are rough instruments, and their origins should be declared. The purchase figures describe households buying cars and computers in the late 1980s. None of it is B2B software, where a budget holder and a security review give a sincere yes still more places to stall. Treat the fractions as a starting discount. The best correction is your own: if you have asked a “would you buy” question before, compare the answers with what happened, and use that ratio from then on.
One trap is worth knowing about. Asking the question changes the people you ask. In a study of car and computer buyers, those who were asked once about their plans went on to buy at a higher rate than those who were not. So do not check your forecast against the customers you surveyed. Measure it on people you never asked.
What already happened
A discount makes a weak question less dangerous. It is better to ask a stronger one.
Rob Fitzpatrick’s book The Mom Test made the rule popular among founders: ask people about their lives and what they have already done, and leave your idea out of it. The logic follows from the research. A prediction costs nothing to give. The past has already happened, and there is nothing left to be optimistic about.
So “Would you use a feature that does this?” becomes “Tell me about the last time you needed to do this. What did you do?” “Would you pay for it?” becomes “What have you paid for, or built, to deal with this so far?” “How much would you pay?” becomes “What do you spend on it today, in money and in hours a week?” And “Would your team adopt it?” becomes “Who else would need to agree, and what happened the last time you brought in a new tool?”
Two of the answers are worth more than the rest. What a customer spends today is a price floor you can verify, whereas “how much would you pay” is a ceiling you cannot. And a workaround is the strongest evidence of demand available without building anything. Somebody who has made their own spreadsheet, or who pays for a second tool to fill the hole, has already spent real effort on the problem, and that deserves weight when you prioritize feature requests.
The weakness of questions about the past is memory, so keep them recent and specific. How far back it is safe to ask is the subject of recall bias in surveys.
Something that costs
The cleaning-product experiment has one more thing to teach. The answers that came closest to real sales were those given when people knew they might have to buy. You can borrow the effect by ending every conversation with a request that costs the customer something.
The strongest is money: a pre-order, a deposit, a paid pilot. Next comes a signature, such as a pilot agreement with a start date. After that, reputation, in the form of an introduction to their manager or to whoever holds the budget. Then time: a second call, access to their data, an hour with a prototype. Last, and worth nothing, come words. “That sounds great, keep me posted” is a polite no. The ranking is our own reasoning, since nobody has tested it for software, but the principle beneath it is the one the research supports. A commitment has a price. A compliment does not.
The 120 customers who said yes to the add-on were not promising to buy it. They were saying that it sounded reasonable, on the day, at no cost to themselves. Read that way, the survey was entirely accurate.
So when a slide next announces that 30% would buy, halve the number, or cut it to a third if the idea is new, and shave the price. Then go and find out what those customers did the last time they had the problem, and what they will commit to today. Count the commitments. They are the only part of the answer that anybody paid for.
How Diaform helps
Diaform collects what people say, so the say-do gap applies to it as it does to any survey. It will not close the gap. What it does is make “last time” questions practical with many customers at once: you write them as open questions and set how far the AI interviewer should follow up, and it asks for specifics when an answer is vague. Each conversation keeps word-for-word quotes and the full transcript, so you can tell a real workaround from a compliment. There is a 14-day free trial.
