
Validation & Market Fit
Compliments are not a purchase order.
People can like an idea, forward it, and even leave an email, without ever changing their behaviour enough to keep a company alive. The useful evidence is a costly commitment, read against a bar you set before the conversations went well.
Marcus Startup Advisor18 min read
Contents
A kind sentence is information, and it is not an order
Ask a thoughtful person whether your idea is interesting and you will often hear that it is. The sentence is not worthless. It tells you the explanation was clear enough to earn a polite response, and sometimes it tells you where the explanation snagged. It does not tell you that the person will spend money, switch a workflow, or defend the purchase to someone else. Treating the kind sentence as a purchase order is how early pipelines fill up with ghosts.
The pressure to do that is ordinary. Founders are looking for a reason to continue, and listeners are looking for a way to be decent. Those motives can line up and produce a meeting that feels like demand. The meeting may even contain a useful fact. The error is the upgrade in status. “She said she would use it” becomes “we have a customer”, and the hiring plan or the raise moves forward on the upgraded sentence. The upgrade is an inference. It should be labelled as one, in the sense argued in What is the company trying to prove next?.
This article is about that upgrade. The reader is a founder, or an adviser, deciding whether a run of positive conversations justifies the next spend. The intent behind the piece is the question people actually ask each other: what counts as validation, how to test willingness to pay, and how to tell a compliment from a commitment. No search-volume figure is attached to those phrases. This draft did not use Search Console or Keyword Planner data for the site, and it does not invent one. The topic is here because the mistake is expensive, not because a tool marked it as trending.
Two kinds of evidence carry the argument. The first is a founder’s contemporaneous account of how Buffer moved from a page, to a pricing click, to a charge. The second is a peer-reviewed meta-analysis on the gap between what people say they would pay and what they pay when the money is real. Neither source is a script you can paste onto every business. Both are better than another round of encouragement.
What Gascoigne says he did, in his words and his limits
On 16 February 2011 Joel Gascoigne, founder of Buffer, published Idea to Paying Customers in 7 Weeks. It is the first post in a series he called Building Buffer. He is reporting his own experience, in the first person, so that other people can compare it with theirs. It is not an audited customer study. Where he is vague, the vagueness stays. Filling it with counts from memory, or from a later retelling, would turn a primary account into fan fiction.
The problem he started from is concrete. He had been using other tools to schedule Twitter posts, and he still found himself sending several at once while reading the news in the morning. He mentions flooding people with five tweets. He wanted a queue, so that he did not have to pick a time for every post. He had suggested the feature to existing products. They had not built it. He decided to build it, and he says he began coding before he had tested whether anyone else wanted it. When he noticed that, he stopped. The sentence he quotes from Eric Ries, as Ries’s answer to how minimal a first version should be, is “Probably much more minimum than you think.” This article has not re-read Ries’s guide in order to certify the quotation. It is Gascoigne’s quotation, inside Gascoigne’s post, and it is cited as such.
The first public test was two pages. He tweeted the link and asked what people thought. A few people left an email, and he received feedback by email and on Twitter. He decided that this was validation, and he uses Ries’s phrase “validated learning” for it. He is entitled to describe his own bar. A reader is entitled to notice how low it was. “A few” is his quantifier. There is no denominator. A few enthusiastic replies from people who already follow you can be a reason to keep investigating. They are not a market.
He then did the more interesting thing. He put a pricing page between the two existing pages, so that a visitor had to pass the price before leaving an email for a launch note. He says this extra click tested which plan people selected and, because it was an extra click, tested whether they were keen. The result, in his words, was that people were still clicking through and leaving emails, and that “a small number of people were clicking on paid plans.” Again he publishes no count. Again nobody had paid. The click was a stated preference for a plan on a page that could not charge them. He treated it as enough to start building the real product. That is his decision, recorded honestly. It is not a result you can borrow as “pricing was validated”.
The build took longer than the story he had told. There was a Hacker News thread about a November startup sprint, with people trying to launch by the end of the month. He had told people the first version would take a week. He gave himself the end of November as a cut-off and built in evenings and weekends for seven weeks. He left out a guided sign-up that he had thought was vital, because the date arrived. Buffer went live on 30 November. The post itself does not write the year next to that date. It is dated 16 February 2011, and he says the learning since launch had covered about two and a half months. The November of the launch is therefore 2010 by that arithmetic. The inference is labelled here so it is not smuggled in as his footnote.
He had expected a long grind. He had a previous product where he had not asked enough people whether the problem was theirs. He launched Buffer, he says, embarrassed by how rough it was, and ready to change it. The signal he treats as commercial arrived faster than that posture. He had his first paying customer within four days. He calls that payment “a strong signal that the solution provided enough value to build a viable business.” After it, he stopped himself adding features and turned toward marketing and more customer development. By the day of the post he reports over 500 users, “many of whom are active”, and revenue on paid monthly plans “at a conversion rate of around 4% of people upgrading.” He also says the team later changed pricing plans. He does not say what the prices were. A retelling that adds a five-dollar plan, or a precise email-to-paid funnel, is not this post.
| Step in his account | What the post actually reports |
|---|---|
| Two-page page, link shared on Twitter | A few people left an email and some feedback. He called that his first validated learning. He does not give the count. |
| A pricing page inserted before the email | People still left emails, and a small number clicked a paid plan. He does not give the count. Nobody had been charged. |
| Build, evenings and weekends | He had told people the first version would take a week. It took seven weeks. He left out features he had thought were vital, including a guided sign-up. |
| Launch on 30 November | The post is dated 16 February 2011 and describes about two and a half months since launch, so the November in question is 2010. He does not write the year in the date itself. |
| Within four days of launch | The first paying customer, on what he calls a rough product. He treats that payment as a stronger signal than the earlier clicks. |
| By 16 February 2011 | Over 500 users, and about 4 percent of people upgrading to a paid monthly plan. He also says pricing was later changed. He does not state the prices. |
Each row is taken from his 16 February 2011 account. Where he gives no count, the cell says so. This is not a conversion chart, because a chart would require numbers the post does not contain. Source: Joel Gascoigne, Buffer, 16 February 2011.
The table is the right figure because the post is a sequence of judgements, not a series of comparable rates. A bar chart of “a few”, “a small number”, “one paying customer” and “about 4 percent” would pretend those phrases share a unit. They do not. The 4 percent is a later description of upgrades among users of the live product. It is not the click-through rate of the pricing page. Mixing them is how a careful account becomes a growth slide.
What his sequence can teach, and what it cannot
Read as a founder’s record, the post supports a few claims and refuses several others. He caught himself building before a test, and he stopped. He distinguished, in practice if not in a formal memo, between “people will leave an email” and “people will click a paid plan” and, later, “a person will pay”. He regarded the payment, four days after a rough launch, as a different class of signal from the earlier clicks. He used a public deadline to ship without the features he had been tempted to call essential. He then, by his own account, resisted the developer’s instinct to answer good news with more product, and spent effort on distribution and on talking to users.
The post does not support a method called “launch in seven weeks”. Seven weeks is how long this version took him, in evenings and weekends, after he had underestimated it as one week. It does not tell you that a two-page site is sufficient evidence to hire or to raise. He did not hire on the back of the emails. He built a narrow product, and he treated the first charge as the moment the commercial question changed. It does not publish a price, so it cannot validate any particular price. It does not say that 4 percent is a good or a bad upgrade rate. “Around 4%” is a description of Buffer at that date, in his words, among “over 500 users”. It is not a benchmark. A team that copies the percentage and ignores the product, the audience and the era is copying a costume.
It is also one founder’s memory, published a couple of months after launch, by the person with the strongest interest in the story coming out well. That does not make it false. It means a sceptical reader keeps his quantifiers. “A few.” “A small number.” “Around 4%.” The value of the post is that he did not invent precision he did not have. The value of using it in your own company is the ladder, not the brand. Email, then a price shown before the email, then a charge, then a renewal you have not yet earned the right to assume. Buffer’s post stops at the early charge and the early upgrade rate. Renewal, cost of service and margin are the next questions, and they are the subject of A useful product is not automatically a viable company.
Stated prices and paid prices are not the same measurement
Gascoigne’s pricing page asked people what they would choose when the choice had no cash consequence. That is a stated preference. The charge four days after launch was a revealed one. Economics and marketing research have spent decades on the gap between the two, and the honest way to use that work is to quote a study rather than to wave at “everyone knows surveys lie”.
In the Journal of the Academy of Marketing Science in 2020, Jonas Schmidt and Tammo H. A. Bijmolt published a meta-analysis of hypothetical bias in willingness to pay for consumer goods. The paper is here. They define the bias as the difference between what people say they would pay in a hypothetical setting and what they pay when the payment is real. Across 115 effect sizes, from 77 studies reported in 47 papers, they find that the hypothetical bias is about 21 percent on average. They also report a result that cuts against a common piece of folk wisdom in research teams. Indirect methods of asking, in their analysis, overstated real willingness to pay by more than direct methods did. The abstract’s point is that the convenient assumption, “indirect is safer”, was not what these studies showed.
This article does not redraw their funnel plot or their forest of effect sizes. The individual study results needed to draw those figures accurately are not reproduced here, and a decorative chart with a single 21 percent bar would look more precise than the evidence being offered. The number to carry into a founder conversation is the authors’ average for the body of consumer-goods studies they assembled, together with the limits on that average. It is not a measured result for SaaS or B2B purchasing decisions. It is not a correction factor you can subtract from a founder interview. It is not a finding about enterprise procurement. It does not say that every stated price is 21 percent too high, or that a direct question is free of bias. It says that, in that literature, hypothetical payment and real payment systematically diverge, and that the size and direction of the gap depend on how the question is asked.
The practical implication is modest and strict. A conversation about what someone might pay is a way to find the shape of the offer and the objections. It is a poor way to book revenue. If the decision you are about to take needs a price, put the price in a commitment that costs the buyer something, even if the something is small, dated and reversible. Then record who took it, who refused, and the reason you were actually given, which may not be the reason you preferred. A small real test beats a large hypothetical one when the next spend is real. A large hypothetical study can still be the right tool for a question you cannot yet put to market. The mistake is using the cheap tool and describing it with the authority of the expensive one.
Ask about the last time, not about a hypothetical self
“Would you use this?” invites a person to invent a future version of themselves and then to be nice to you. “Tell me about the last time this problem showed up” invites a memory. The memory can be wrong, edited and self-serving. It is still anchored to an event, a cost and a workaround. You can ask what they did, who else was involved, what it cost, and what they had already tried. You can also notice when the event was two years ago. A problem that last occurred two years ago is a weak reason to buy software this quarter.
The current alternative is the real competitor. It may be a spreadsheet, a colleague, a weekly meeting, or doing nothing. A clumsy alternative can survive because replacing it risks a day of disruption for a pain that is only annoying. Your job in the conversation is to find out whether the pain is expensive enough, often enough, for this buyer, to beat the switching cost. Enthusiasm for your demo does not answer that. A story about last Tuesday might.
Listen for a buyer as carefully as you listen for a user. The person who feels the pain may not be allowed to pay. A practitioner’s compliment and a practice manager’s budget are different facts. If you leave the conversation without knowing who would have to say yes, you have a user story and a missing commercial role. That missing role is the next conversation, not a reason to average the compliments into a market.
Refusals deserve the same notebook as the yeses. A no can mean the wrong person, the wrong week, a muddy offer, or a problem that does not hurt. You usually cannot tell from one conversation. You can tell when the same reason arrives from the fifth similar buyer. Design the next test around the reason that is still plausible, and retire the reason the conversations have killed. Do not collect more compliments in the hope that the refusals were a sampling error. Sometimes they were. The way to find out is a better sample of the same question, not a softer question.
A labelled hypothetical, so the bar is visible
The company in this section is invented. It is here because Gascoigne’s post does not give counts, and inventing counts for him would be worse than a clearly fake example. Hale is a two-person team building a tool that turns a project manager’s status meeting into a written update. They speak to twelve project managers. Ten say the meeting is a waste. Six say they would “definitely use” a tool like Hale’s. Three ask to be told when it launches. One offers to introduce Hale to a colleague. Nobody is shown a price. Hale’s note to itself says “10/12 validation, strong pull.”
That fraction is a compliment rate. It is not a demand rate. A decision memo that deserves the name would split the lines. Observation: ten of twelve described the meeting as wasteful, and three asked for a launch note. Inference: these managers will pay, and their employers will let them. Unsupported dependency: no price has been offered, and no manager has been asked to use a draft on a real meeting this week. Threshold, written before the next twelve conversations: five managers, in companies of a stated size, complete one real meeting with the draft and the buyer agrees a paid month at a stated price. Anything short of that, including a warmer set of compliments, does not authorise a hire.
The five is not a research finding about project management. It is an illustration of a bar that can be missed. You should set your own against the cost of the commitment you are protecting. If the commitment is “keep talking to people for another fortnight”, a handful of serious conversations can be enough. If the commitment is “hire, raise, and rewrite the product”, the bar has to be something a sceptical co-founder would accept as costly to the customer. Gascoigne’s own ladder is a useful comparison. He did not hire on the emails. He built, he charged, and only then did he treat the commercial signal as strong. Teams that invert that order are not “doing Buffer”. They are doing the part of the story he had already decided was too weak to build on, and they are skipping the part he treated as the signal.
A gate you can fail
Write the gate down before the conversations, not after you have seen who was friendly. Four lines are enough.
The segment. Who you will talk to, and who does not count even if they are encouraging. A friend in the industry can improve the offer. A friend is not the segment unless friends are the buyers.
The commitment you will ask for. A dated pilot, a payment, a data export, an introduction to the economic buyer, or another act that costs the other person something they would not spend on a stranger’s idea. “Leave your email” can be a first rung, as it was for Gascoigne, if you describe it as a first rung. It cannot be the only rung and also the bar for a raise.
The threshold. How many of those commitments, by when, would change your decision. Include the quality, not just the count. Five pilots inside one company can be one buyer’s enthusiasm. Five pilots in five companies of the same type are a different observation. Write which one you mean.
The cost limit. The time and money you will spend to find out, and the decision if you finish the window short of the threshold. Continue, change the offer, or stop. “Learn more” is allowed only if you name the missing fact. Otherwise it is a way to avoid the stop.
Use the gate in the conversation itself. Tell the buyer what you are asking them to do, and what you will not pretend they have done. People can smell a fake pipeline. They are more willing to give a real no, or a real yes, when the question is specific. Specific questions also protect you from your own memory. A week later, “she seemed keen” expands. A note that says “she would not book a pilot and asked to be emailed in the autumn” does not.
When the window closes, read the notes against the gate rather than against your hopes. Proceed only if the commitments you defined have actually arrived. Change the offer if the problem keeps showing up in the last-time stories but the commitment does not. Stop the path if the stories themselves are thin. Then, if someone has paid, move to the question the payment does not answer, which is whether you can deliver at a cost the price can carry. Payment is necessary and not sufficient. That is an argument from the structure of a business, not a claim that Buffer’s fourth day proved a durable company. Gascoigne does not claim that in the post. He claims a strong early signal, and then describes the work still in front of him.
What to do with the next ten conversations
Before you book them, write the segment, the commitment, the threshold and the stop. Put the note where a co-founder will see it. A gate that lives in your head will be edited by the third friendly call.
In the conversation, spend the time on the last real episode and on the current workaround. Ask what it cost and who else felt it. Show the offer only after you understand that episode, and show a price if the decision you are feeding needs a price. A demo without a price teaches you about the demo. A price without a commitment teaches you about politeness. The commitment, even a small one, is the measurement Schmidt and Bijmolt’s literature says the hypothetical question is not.
Keep a column for what you will not count. Compliments. Follows. “Keep me posted.” A senior person saying the problem is important and then leaving the decision to someone you never meet. Those can be leads to a further question. They are not yeses. Gascoigne’s “small number” of pricing clicks would sit in a middle column: more than a compliment, less than a charge. His first paying customer would move columns. If your notes do not have a middle column, you will either over-count the clicks or throw them away. Both are worse than labelling them.
After ten conversations, update the memo rather than the slogan. If you have the commitments, make only the decision the gate allowed. If you have last-time stories and no commitments, change the ask or the buyer before you change the company. If you have neither, stop spending the month on this path. The aim is not a perfect study. The aim is a record that can survive a sceptical reading on the day you are tempted to hire.
If you want that reading done with you, bring the notes, not the highlights. The validation stage is the short checklist. A Focused Session is for the gate and the spend behind it. The result worth having is a smaller claim than “the market wants this”. It is a named buyer, a costly next step, a number you can miss, and a clean distinction between what people said to be kind and what they did when kindness was no longer enough.
Sources
- Joel Gascoigne, Idea to Paying Customers in 7 Weeks: How We Did It, Buffer, 16 February 2011 The founder’s own account of the two-page test, the later pricing page, the 30 November launch and the first paying customer. It is not an audit.
- Jonas Schmidt and Tammo H. A. Bijmolt, Accurately measuring willingness to pay for consumer goods: a meta-analysis of the hypothetical bias, Journal of the Academy of Marketing Science, 2020 Reports an average hypothetical bias of about 21 percent across 115 effect sizes from 77 studies in 47 papers. This is a finding about those consumer-goods studies, not a law for every founder interview.
- Marcus Startup Advisor, What is the company trying to prove next? How to choose the commitment the evidence is supposed to justify.
- Marcus Startup Advisor, A useful product is not automatically a viable company What to examine once someone has actually paid.