How to read early signs of product-market fit

1 The bottom line

Early product-market fit shows up as strong feeling, then unprompted pull, then retention and word of mouth. Read those in order, and if they stay weak, fix the product or pivot before spending on growth.

2 Key takeaways

• Start with how people react, not with dashboards. Early numbers are too small to mean anything.
  • Early on, eyes lighting up beats DAU, MAU or D7 retention, which are too thin at that stage Josh Woodward, Google Labs15:46.
  • A Gemini feature was pulled before launch after in-person tests showed the reaction was not there Josh Woodward, Google Labs15:46.
• Trust pull that nobody asked for. People who act on their own care about the product.
  • After a weekend test of a half-baked tool, friends asked to use it. He had only ever felt push before Dmitri, YourMove AI5:40.
  • Around January, 5 to 10 people a day posted unprompted about loving Flow, many who had never posted about a product Tanay Kothari, Wispr Flow34:25.
• Put a number on fit with the "very disappointed" survey. A shared target ends arguments about when to scale.
  • Companies that struggled to grow almost always had under 40% very disappointed Rahul Vohra, Superhuman55:35.
• Do not believe early buzz until cohorts confirm it. Small samples and cheap traffic flatter a weak product.
  • Word-of-mouth signups with no paid acquisition felt like fit, but cohorts showed a leaky bucket Sam Chaudhary, ClassDojo15:42.
  • At about 90 downloads a month, TimeClock's funnel looked great while margin stayed under 1 percent Jordan Bryant, TimeClock2:58.
• Hold growth spending until the signals are there, then keep going, fix or pivot. Scaling a weak product only makes a bigger mess.
  • Visible delayed growth experiments until the product retained people, then went from $1M to $10M subscription ARR in two years Luke Martin-Fuller, Visible37:13.
  • Airbnb never ran out of growth ideas, which is a reason to stay the course Dalton Caldwell, Y Combinator18:16.

do people use it? Do they like it? Do they come back? Do they talk about it?

Ami Vora and Mike Krieger, Anthropic

2.1 Read the first reactions

Before the numbers mean anything, watch how real people respond to the product.

• Put the prototype in people's hands and watch their faces. Source: Josh Woodward, Google Labs15:46
  • Take it outside the building. If people lean in, you have hit a real pain point.
  • Steve's camera app got a polite reaction to tap-anywhere-to-shoot, but a strong one when shaking the phone flipped the camera Steve P. Young, App Masters9:32.
  • He then led all his marketing with the shake feature. It earned press on The Next Web and strong sales Steve P. Young, App Masters9:32.
• Look for a few people who love it, and ignore small lifts. Source: Claire Butler, Figma23:40
  • With tiny numbers, a 5% email lift says nothing about whether the product works.
  • Designers grabbed the laptop out of Dylan's hands during demos, and some teams used Figma full time. Figma left stealth on that signal.
• Ship weekly to a small beta group and note which moments feel good. Source: Ajay Mehta, Tolan11:03
  • Ship on Friday, test over the weekend with friends and TestFlight users, and start fixes on Monday.
  • Note which interactions leave people feeling good and which bore them.
  • They did this for about 6 months before launch. Even janky early builds with long response times and a static image showed inklings of emotional resonance.
• Expect indifference, and treat confusion as a clue. Source: Gustaf Alströmer, Y Combinator30:00
  • The most common reaction is not hatred but indifference. Most people sign up and forget.
  • Reach out again in two weeks, six months or a year, especially after improvements.
  • If many users do not understand the app, you may target the wrong users, miss the right people, or have unclear usage. Improve features, clarify use and add FAQs Christine Dart, Helpshift9:17.

2.2 Look for pull you did not ask for

The strongest early signs are things people do without being prompted.

• Count people who share it before you market it. Source: Maor Shlomo, Base4441:04
  • Start with a few close friends, sit with them and fix what breaks.
  • Invest in marketing only when users you do not know show up through word of mouth. Before that, you run a revolving door.
  • He began with 3 friends. After the 11th user came from outside his circle, he launched and invested in marketing.
• Treat loud complaints as a good sign and silence as a warning. Source: Gaurav Misra, Captions19:16
  • Users who ask for missing features care enough to want it better. Use the loudest complaint to pick what to ship next.
  • Angry emails during an outage show the same thing. Many writers say they love the team once it is fixed Tanay Kothari, Wispr Flow34:25.
  • Jeff's SQL startup was down for 10 to 20 minutes and customers only murmured. He missed that this meant no fit, and spent many more years on it Jeff Weinstein, Stripe18:00.
• Read breakage and demands to pay as proof of need. Source: Jon Noronha, Gamma16:50
  • Two to three weeks after launch, Gamma was still unsure. Then support volume exploded and over half of messages asked how to pay for more AI.
  • Intercom went from 3 messages a day to 50 and then 300. Cash flow turned positive in three months.
  • TBH had to be geofenced to keep its servers up. In consumer apps, if there is any uncertainty, it is not working Nikita Bier, TBH and Gas36:22.
• Watch where users go when the product is wrong for them. Source: Andrew Ambrosino, OpenAI54:09
  • Marketing, comms, finance and legal staff used the Codex app even though it showed code and asked for approvals. They would not leave it for apps built for them. That was the signal to generalize Codex.
  • Employees using personal Gmail accounts for NotebookLM led companies to ask to make it official with work email Raiza Martin, NotebookLM25:35.
  • For a social product, heavy internal use is a strong sign, and no internal buzz is a warning Kevin Weil, OpenAI7:13.

2.3 Put a number on it

Once a few people use it, measure how much they would miss it.

• Ask the "how would you feel if you could no longer use this" question. Source: Rahul Vohra, Superhuman55:35
  • Offer three answers: very, somewhat or not disappointed. Use the share of "very" as your score. Tavel prefers it to NPS Sarah Tavel, Pinterest57:45.
  • Ask only people who used the product two or more times in the last week or two. Skip people who only signed up or saw a demo Sean Ellis22:38.
  • Ellis needed at least 30 responses. With 10 people and 4 very disappointed, you get a signal but not a sample to go to market on.
• Set the bar before you start. Source: Sean Ellis17:58
  • The 40% bar came from patterns across startups and is not exact. Its power is a shared target for when to grow.
  • Nubank uses 50% and rarely scales a project before it clears the bar Jag Duggal, Nubank16:17.
• If the score is borderline, cut it by behavior and build for the bullseye group. Source: Jag Duggal, Nubank23:23
  • Find who loves it most, then make it easier for everyone else to reach that same behavior.
  • Nubank's payments assistant scored about 40% overall but 70% among users with 4+ commitments on at least two of four payment rails. It grew from hundreds of thousands to over 10 million monthly actives in about 15 months.
  • Each planning cycle, spend half your time on what people love and half on the objections of the "somewhat disappointed" Rahul Vohra, Superhuman57:58.
• Check that the product shortens the customer's work. Source: Oji Udezue, Typeform28:55
  • Draw the workflow before and after your software and measure both. If the new one is 2X or 3X shorter, the problem is sharp.
  • People do not notice a 20% improvement. They feel it and pay at about a 3X acceleration Oji Udezue, Typeform13:07.

2.4 Confirm with retention and word of mouth

Real fit means people come back and tell others without being paid to.

• Check whether engaged users stick around, then ask what they tell friends. Source: Seth Godin30:01
  • Do not call it fit on signups alone. Look at cohorts.
  • ClassDojo's retention was mediocre until its multiplayer jump fixed it. Sam says the retention analysis was among the most important the company has done Sam Chaudhary, ClassDojo15:42.
  • In a marketplace, better cohorts and new participants arriving on both sides "without a fight" show tipping Sarah Tavel, Pinterest1:20:29.
• Look for word of mouth you did not pay for. Source: Julia Schottenstein, dbt Labs11:37
  • Ask whether users can stop talking about it and want to share it with other teams.
  • Mercor had no one in sales and marketing for a year and a half. Over half of its marketplace comes from referrals Brendan Foody, Mercor40:30.
  • About 20% of Skylight's respondents say it is life-changing. The aim is to push that toward 50%, since these people bring in new customers Michael Segal and Mark Ungerer, Skylight41:07.
• Be far better on the key lever if you want people to recommend you. Source: Nilan Peiris, Wise13:50
  • Marginal improvements earn usage, not recommendation.
  • Wise at 5.9% against a 6% incumbent got use but no talk. Advocacy began at eight to 10 times cheaper.
• Discount signals that are cheap to fake. Source: Gustaf Alströmer, Y Combinator27:05
  • YC acceptance, investor checks and press make founders feel they have fit when they do not.
  • At a few hundred dollars a day, ad algorithms find the easy buyers. Costs go out of control near $10,000 a day Vivek Girotra, Times Internet10:50.
  • A launch spike in a new market can be everyone who could subscribe being pulled forward Crystal Widjaja, Gojek and Kumu26:40.
  • Emailing all members about a new feature inflated engagement. Now they ship quietly and announce only if there is signal Ethan Ethier, Built With Science44:48.

2.5 Decide: keep going, fix or pivot

If the signals are weak, work out why before you spend on growth or change direction.

• Do not spend on marketing or conversion until the product earns it. Source: Michael Segal and Mark Ungerer, Skylight39:14
  • Below about 40 NPS, marketing gives a false negative, because you cannot tell whether marketing or product is failing.
  • Growing a product without fit gives users a bad experience, and people rarely try something twice Luc Levesque, Shopify18:59.
  • Skylight pushed marketing for its calendar in 2021 and 2022 and it did not work. It took off in fall 2022, after a rebuild.
• Diagnose a lukewarm reaction before you quit. Source: Mike Maples, Jr., Floodgate42:57
  • Either the insight is wrong (stop), the implementation is wrong (iterate), or you are asking the wrong customer.
  • Okta first built cloud systems management and heard "meh". It asked about top priorities, learned it was identity management, and then it worked.
• Check progress and your remaining ideas. Source: Dalton Caldwell, Y Combinator18:16
  • List the good growth ideas you have not tried. With half a dozen or more, try them before pivoting. If you are down to paying influencers, you are out of gas.
  • Good signs are news every week or two, such as customers bought or new pricing Gustaf Alströmer, Y Combinator47:00.
  • A metric moving from 10% to 10.5% to 10.6% is up but may not be on a path to what the business needs Eric Ries, Long-Term Stock Exchange1:36:15.
• If you pivot, commit fully and move toward what you know. Source: Yuriy Timen, Grammarly46:25
  • Splitting effort is a marginal correction, not a pivot. Half measures fail either way.
  • A good pivot "is like going home". Brex and Retool both pivoted toward domains the founders knew Dalton Caldwell, Y Combinator15:35.
  • Change persona, problem and promise first, since that is cheaper than rebuilding the product Todd Jackson, First Round Capital32:03.
  • Josh Woodward expects three to five pivots, so fall in love with the problem, not the solution Josh Woodward, Google Labs16:40.
  • Ask whether you would start this company again today. If not, you are free to shut down or pivot hard Eric Ries, Long-Term Stock Exchange58:00.

2.6 Where they disagree

Experts differ on whether fit is a clear moment, how much to trust excitement, and which survey bar to use.

• Fit is obvious and binary, or a slow ramp.
  • In consumer apps, if you have to wonder, you do not have it Nikita Bier, TBH and Gas36:22.
  • For Notion it was "a very gradual ramp" Ivan Zhao, Notion25:40. At Learneo, nobody put up a sign. Several milestones landed together Andrew Grauer, Learneo16:51.
  • YNAB sat in the store for six years before it clicked in 2021 Steve P. Young, YNAB2:45.
• Excitement is the metric, or it is noise.
  • Woodward says eyes are "the metric" early on Josh Woodward, Google Labs15:46.
  • Udezue is "a little over" that reaction, since excitement can be noise. He trusts measured workflow compression more Oji Udezue, Typeform30:45.
• The survey bar depends on the product and the culture.
  • 30% may be enough in a pessimistic market like Hungary, while Nubank asks for 50% in Brazil Sean Ellis17:58.
  • High scores can come from switching costs. Webs.com scored about 90% in a commoditized category because users had invested so much in their sites Sean Ellis19:20.
  • Customer love is not enough. WeWork and Casper had demand and satisfaction but never efficiency Todd Jackson, First Round Capital17:29.
• Whether to skip the testing process.
  • If you have evidence the product clicks and you have conviction, run with it. Base44 sold for $80 million in six months without a formal process, but they call that selection bias Jake Knapp and John Zeratsky, Character Capital1:08:20.
  • Conviction can come later. Build it from customer signals, not before them Dalton Caldwell, Y Combinator51:10.