Before you write production code, write the demand bet as a number: at least X% of a named group will take a paid action at a stated price, by a date. Alberto Savoia calls this the XYZ Hypothesis. Then interview buyers about the last time the problem hit them, with your idea kept out of the call. Abbie Griffin and John Hauser found that 20–30 interviews in one segment surface 90% or more of customer needs. Ask each strong prospect for something that costs them: a next meeting, an intro to the budget owner, or money. Write the kill line before you run the test, and build when paid commitments clear it. In CB Insights’ 2026 review of 431 VC-backed shutdowns, poor product-market fit appears in 43% of the 385 cases with a known cause.
This page covers the work before the first line of production code. Once a buyer wants proof in their own environment, read how to convert a pilot to a paid contract. Once they want to sign, read how to negotiate your first customer contract.
Rank demand signals by what they cost the buyer
| Signal | What the buyer gives up | What it proves | Source |
|---|---|---|---|
| Compliment or “I would buy that” | Nothing | Nothing. Read it as a polite exit | Fitzpatrick; Alströmer |
| Survey answer | A few minutes | Stated interest, shaped by your question | Alströmer |
| Email or waitlist signup | An email address | Initial interest in the offer as written | Savoia; Dropbox |
| Next meeting with a set agenda | Time | The problem matters enough to spend more time on it | Fitzpatrick |
| Intro to a peer, the team, or the boss | Reputation | Your contact will stake standing on it | Fitzpatrick |
| Letter of intent | A signature with no legal obligation | Intent at a price, if the price is on the page | Fitzpatrick |
| Pre-order or deposit | Money | Willingness to pay before delivery | Fitzpatrick; Savoia |
| Paid pilot | Money, time, and reputation | Willingness to pay plus an owner inside the buyer | Fitzpatrick |
Fitzpatrick’s rule of thumb in The Mom Test: “The more they’re giving up, the more seriously you can take what they’re saying.” Alströmer, a YC group partner, says a five-minute video interview teaches you more than 500 or 5,000 survey responses. Paul Graham describes the trap with a made-up social network for pet owners: friends say “Yeah, maybe I could see using something like that,” and summed across the population that reaction gives you zero users.
The FounderNexus session view: money is the signal, and interviews or surveys count for little on their own. While you test, state the full price in writing and offer the early buyer a lower test price. Their reaction to the full number is the data. Judgment from a FounderNexus session, not a survey.
Write the bet as a number first
Savoia, Google’s first engineering director and the author of The Right It, wrote the format on a Stanford whiteboard in 2016: “At least X % of Y will do Z.” X is a share of the target market, Y names the market, and Z is the action you expect. His students’ version: “At least 10% of people who live in cities with an AQI level greater than 100 will buy a $120 portable pollution sensor.” Writing the number exposed a disagreement. One student priced the device at $200 and another at $80 to $100.
Then zoom in. Savoia’s tactic is to think globally and test locally: turn the market-wide bet into a smaller one you can test this month on people you can reach.
| Part | Market-wide bet | Local test |
|---|---|---|
| X | Share of the market that must act | Share of the people you contact |
| Y | The full segment | The segment members inside your network or one channel |
| Z | Buys at the target price | Pays a deposit, signs a priced pilot, or books a second call with the budget owner |
Editorial example, not data: “At least 10% of revenue operations leads at B2B SaaS companies with 50 to 500 employees will pay $2,000 a month for renewal forecasting within 60 days of a first demo.” The local version: “At least 3 of the 30 revenue operations leads we can reach this month will sign a paid 60-day pilot at that price.”
FounderNexus session judgment on Z: for long sales cycles, put a time bound in the sentence. A prospect who takes months to answer fails the bet on time even if the interest sounds real. Test demand first. Whether you can build it, and at what cost, comes after.
Run problem interviews before you show the product
Migicovsky, who founded Pebble, built his Startup School lecture on The Mom Test’s three rules: talk about their life instead of your idea, ask for specifics instead of hypotheticals, and listen more than you talk. Alströmer adds that you should introduce the idea at the end of the call or not at all.
| Ask | Avoid |
|---|---|
| Tell me about the last time that happened. | Would you use a product that did this? |
| What is the hardest part of doing it today? | Which features would make it better? |
| What have you tried, and what don’t you love about it? | What would a better product look like to you? |
| How often does it happen, and what does it cost you? | Questions with a yes or no answer |
| Who has the budget and the authority to fix it? | Two questions in one breath |
The left column comes from Migicovsky’s five questions and his prototype-stage checks on cost, frequency, and budget. The right column is Alströmer’s list of questions that derail an interview.
Talk to earlyvangelists. Steve Blank’s profile of the buyer who takes a risk on an unfinished product: they have the problem, know they have it, search for a fix on a timetable, have built a workaround, and have or can get the money. Alströmer’s test for value: are people paying for other solutions today? Spreadsheets compete with hundreds of startups, so a new product has to beat them by a wide margin.
Plan the count. Griffin and Hauser’s 1993 Marketing Science study concluded that 20–30 interviews should identify 90% or more of customer needs in a relatively homogeneous segment. Alströmer moves to an MVP after 5 or 10 interviews. Use the small number to find the segment and the larger one to map its needs before you commit an engineering team.
Run the calls yourself. Migicovsky: talking to customers is part of the CEO’s job description, and the engineers take part too. FounderNexus session judgment: score each problem on how much it matters to the buyer and how satisfied they are with today’s fix. Look for high importance, low satisfaction, and a short path to money made, money saved, or risk reduced. Book the next conversation before the call ends.
Pick the cheapest test that asks for commitment
| Test | What you build | Fits when | Example (source) |
|---|---|---|---|
| Fake door | A page, button, or ad with the offer and price | Self-serve offers with traffic you can reach | McSpaghetti on the menu before any pasta exists (Savoia) |
| Demo video plus waitlist | A short screen recording of the intended product | The product is hard to build and the audience gathers in one place | Dropbox: 5,000 to 75,000 on the waiting list overnight (Houston, via Eric Ries) |
| Service run by hand | People do the work behind a plain front end | The output matters more than the automation | IBM tested speech-to-text with a hidden typist (Savoia) |
| Deposit or pre-order | An order form and a payment link | Hardware or long-lead products | Tesla asked interested buyers for a $5,000 deposit (Savoia) |
| Short-run offer | A real, limited offer for a few days | Services and marketplaces | Airbnb: 3 guests and $240 on the first night (Savoia) |
| Priced pilot or LOI | Pilot paper or an LOI with price, scope, and dates | B2B buyers with a budget owner | Pilot to paid contract |
Savoia’s note on fake doors: tell visitors at once that the product is not ready, and thank them with a gift close to what they tried to buy. Paul Buchheit’s advice, quoted by Graham, is to “go sell the product ASAP (before wasting time building it).” Founders who do learn whether anyone wants it, and often find the better idea along the way.
FounderNexus session judgment: the cheapest artifact today is often an AI-generated demo built in hours. Put it in front of target buyers and collect your own first-hand data. Industry reports and your own expertise do not count as that data.
Set the kill line before you test
Savoia names three paths for an idea: Drop It, Go for It, or Test It. He warns against the first two when you have no data. After the test, read the result against the bet you wrote.
| Result against the bet | Move |
|---|---|
| Paid commitments meet or beat the X you wrote | Build the smallest version for the buyers who paid |
| Strong pain, little money | Find who owns the budget, or retest at another price |
| Pull from one sub-segment only | Narrow Y and rerun the local test |
| Compliments, no commitments | Treat it as a no. Change the buyer or the problem |
| Interest that takes months to answer | Treat it as a no on the time bound |
There is no reliable public benchmark yet for the share of waitlist signups, LOIs, or fake-door clicks that become paying customers. Set X from your own revenue model and write it down before the first test.
Two more checks from the FounderNexus session. First, find the companies that tried this and died. Call their founders, ask what killed the company, and solve each of those flaws before you build. Second, walk a product-free deck past a few investors with no ask attached. They raise the hard questions early and often name predecessors you missed. Treat their take as data on how the story lands, filtered through each investor’s bias. A design partner who pays nothing does not count as validation, because investors read it as an unfinished product looking for a buyer.
Validate demand in order
Demand validation
- Write the XYZ bet. Percentage, named group, paid action, price, and a date (Savoia).
- Zoom to a local test. People you can reach this month through your network or one channel.
- List earlyvangelists. Buyers with the problem, a workaround, a timetable, and money (Blank).
- Run problem interviews. Past behavior only. Plan 20–30 in the segment (Griffin and Hauser).
- Score the pain. Importance, satisfaction with today’s fix, and the path to money (FounderNexus session).
- Study the predecessors. Call founders of companies that died on this problem (FounderNexus session).
- Build the cheapest artifact. A priced fake door, a demo, or a service you run by hand.
- Ask for a commitment. Time, an intro, or money, with the full price stated (Fitzpatrick).
- Read the result against the kill line. Keep, narrow, reprice, or drop, as written before the test.
- Build for the buyers who paid. Then measure pull with the Sean Ellis survey.
After you build, measure pull
Sean Ellis’s question: “How would you feel if you could no longer use the product?” After benchmarking nearly a hundred startups, he found that companies with strong traction almost always had more than 40% of users answer “very disappointed.” Migicovsky uses the same 40% line in his lecture. Superhuman started at 22% in summer 2017, reached 33% after segmenting to its most enthusiastic users, and hit 58% within three quarters, according to founder Rahul Vohra.
Mistakes that waste the build
Counting compliments. A meeting that ends with praise and no next step is a failed meeting (Fitzpatrick).
Pitching on a discovery call. Your idea biases the answers. Save it for the end of the call (Alströmer).
Asking about the future. “Would you pay for this?” gets a yes that means nothing. Ask what they paid for last time.
Testing without a price. A signup with no price tests curiosity. Put the dollar figure in Z.
Delegating the calls. The founder who will make the build decision runs the interviews (Migicovsky).
Moving the kill line. Rewriting the pass line after the result turns the test into a pitch.
Calling an unpaid design partner validation. Ask design partners to pay something (FounderNexus session).
Pair with pilots, contracts, and messaging
Turn a priced pilot into a purchase order with how to convert a pilot to a paid contract. Write the first order form with how to negotiate your first customer contract. If the first buyers are large companies, check readiness with selling to the enterprise as an early-stage startup. Write the homepage and fake-door copy around the job with how to market an AI product without leading with AI. Bring the evidence into the raise with the seed pitch deck page. Orient the rest of the motion from the GTM hub.
Sources
- Alberto Savoia, Pretotype It, 10th Anniversary Edition. 2022, free PDF, with the bonus chapter excerpted from The Right It (HarperCollins, 2019). Law of Market Failure: most new ideas fail in the market even if competently executed, 80–90% of ideas brought to market; Fake Door pretotype and its ethics; Drop It / Go for It / Test It; think globally, test locally; XYZ Hypothesis “At least X % of Y will do Z” with the Stanford 2016 example (10%, AQI greater than 100, $120; student prices $200 vs $80 to $100).
- Alberto Savoia and Tina Seelig, Quick reference: basic pretotyping techniques. Stanford MS&E 277, Fall 2016. Fake Door (McSpaghetti), Mechanical Turk (IBM speech-to-text with a hidden typist), Impostor (Tesla $5,000 deposit), One-night Stand (Airbnb 3 guests, $80 a night, $240 first night).
- Rob Fitzpatrick, The Mom Test. Chapter five, Commitment and advancement: compliments cost nothing and carry no data; currencies of time, reputation risk, and cash; letter of intent, pre-order, deposit; paid trial with the whole team combines all three.
- Y Combinator, How to talk to users. Gustaf Alströmer, Startup School. Introduce the idea at the end or not at all; a five-minute video interview over 500 or 5,000 survey responses; questions to ask and to avoid; MVP after 5 or 10 interviews; are people paying for other solutions today.
- Y Combinator, Startup School Week 1 Recap: Kevin Hale and Eric Migicovsky. 3 September 2019. Eric Migicovsky, How to Talk to Users. Founders keep a direct connection to users; The Mom Test’s three errors; five interview questions; prototype stage cost, frequency, and budget; 40% “very disappointed” line for product-market fit.
- Abbie Griffin and John R. Hauser, The Voice of the Customer. Marketing Science 12(1), 1993. Interviews with 20–30 customers should identify 90% or more of the customer needs in a relatively homogeneous segment.
- Steve Blank, Perfection By Subtraction: The Minimum Feature Set. 4 March 2010. Earlyvangelist traits: has the problem, knows it, searching on a timetable, built an interim solution, has or can get budget.
- Paul Graham, How to Get Startup Ideas. November 2012. The most common mistake is solving problems no one has; the pet-owner social network reaction; Paul Buchheit on selling before building.
- TechCrunch, How DropBox Started As A Minimal Viable Product. Eric Ries, excerpt from The Lean Startup, 19 October 2011. Drew Houston: the beta waiting list went from 5,000 to 75,000 people overnight after the demo video.
- CB Insights, The top 9 reasons startups fail. 5 March 2026. 431 VC-backed startups that shut down since 2023; reasons identified for 385; poor product-market fit 43%; ran out of capital 70%, usually the final cause.
- First Round Review, How Superhuman Built an Engine to Find Product Market Fit. Rahul Vohra, 13 November 2018. Sean Ellis benchmark across nearly a hundred startups, 40% “very disappointed”; Superhuman 22% to 33% after segmenting to 58% within three quarters.
- FounderNexus. Founder sessions, 2026. Money is the signal; state the full price and offer a lower test price; time bound in Z; demand before buildability; score importance against satisfaction; book the next call; cheapest artifact is an AI-built demo; call founders of failed predecessors; product-free deck past investors with no ask; unpaid design partners are not validation. Not a survey. No closed-session numbers.
Related
- Go-to-market
- How to convert a pilot to a paid contract
- How to negotiate your first customer contract
- Selling to the enterprise as an early-stage startup
- How to market an AI product without leading with AI
- Seed pitch deck: what investors check slide by slide
Founders who have killed or kept an idea on pre-build evidence will pressure-test your demand bet, interview notes, and kill line in a FounderNexus session.