What fit is, and why it comes first

Marc Andreessen gave the term its common meaning in a 2007 essay, The only thing that matters: "Product/market fit means being in a good market with a product that can satisfy that market." He also described how its absence feels: customers are not quite getting value, word of mouth is not spreading, and usage is not growing fast. When fit is there, customers buy as fast as you can serve them.

For a PM, a working definition is more useful. A product has product-market fit with a group of users when a meaningful share of them keep using it at its natural frequency without being pushed, would be upset to lose it, and bring others in. Three parts of that sentence matter.

• A group of users. Fit is never universal. A grocery app can fit busy families in dense neighbourhoods and fail with students in the suburbs. You measure fit per segment, per city, sometimes per district.
• Natural frequency. People buy groceries about weekly, a corner shop restocks two or three times a week, and store staff receive deliveries every shift. Judge use against that rhythm. A grocery app does not need daily users.
• Without being pushed. Use that only happens because of a coupon, a launch campaign or a manager's order tells you little. Fit shows in what people do once the push stops.

Why does this come before growth? Growth spending fills the top of the funnel. Retention decides how much of it stays. Think of each month's new customers as a cohort. If each cohort leaves a lasting layer of regular customers, every dollar of growth adds a layer, and the layers stack up month after month. If each cohort drains away to nothing, growth spending buys visitors who leave, and you have to keep paying forever to stand still.

Dropbox learned this from the other side. In a 2010 talk, its founder Drew Houston showed that paid search had cost the company between $233 and $388 per customer for a product that sold for $99 a year. Yet Dropbox kept growing, through word of mouth and sharing, because the people who used it loved it. One of his slides reads: "Product-market fit cures many sins of management." The reverse is also true: no amount of marketing cures the lack of it.

Fit can also be lost. A competitor arrives, a regulation changes, delivery times slip as you grow. Treat fit as something you keep measuring, segment by segment, every quarter.

The "very disappointed" survey

Retention tells you what people do. A short survey tells you how much they would miss the product, and, more usefully, who misses it most and why. The best-known version comes from Sean Ellis, who coined the term "growth hacker" and helped Dropbox in its early years. It asks one main question:

"How would you feel if you could no longer use [product]?" Very disappointed / Somewhat disappointed / Not disappointed (it isn't really that useful) / N/A, I no longer use it.

Four rules make the answers mean something:

• Ask the right people. Ellis's guidance, as Buffer's co-founder Leo Widrich described it in 2015, is to survey people who have experienced the core of the product, used it at least twice, and used it in the last two weeks. People who tried once with a coupon have nothing to lose yet.
• Get enough answers per segment. About 40 to 50 responses is a commonly quoted minimum for a clear signal. Below that, treat the result as a hint. Drop the N/A answers before you calculate.
• Use 40% as a rough line, not a law. Ellis compared nearly 100 start-ups and found that companies with strong traction almost always had more than 40% of users answering "very disappointed", and struggling ones usually had less.
• Pair it with behaviour. The question asks people to predict a feeling, and people are poor forecasters. A high score with a falling retention curve means something is wrong with one of the two measurements.

The score itself is the least interesting output. The power is in segmenting it. Rahul Vohra, the founder of the email app Superhuman, described how his team did this in a 2018 First Round Review article. In summer 2017, only 22% of their users said they would be very disappointed. The team added three questions to the survey:

1. What type of people do you think would most benefit from the product?
2. What is the main benefit you receive from it?
3. How can we improve it for you?

They then looked only at the users who answered "very disappointed", found which kinds of people they were (founders, managers, executives and business development people), and recalculated the score for those groups alone. It rose to about a third. Next they read what those users named as the main benefit: speed. Finally they split the roadmap in half. One half doubled down on what the very disappointed users loved, such as speed and keyboard shortcuts. The other half fixed what held back the somewhat disappointed users who also valued speed, such as the lack of a mobile app. They set aside feedback from people who would not be disappointed, and from somewhat disappointed users who valued something else. Within three quarters the score reached 58%.

The lesson for any PM: the very disappointed users tell you what the product is for. The somewhat disappointed users who share that main benefit tell you what to fix next. Everyone else is a distraction for now.

Retention that flattens

The survey is a snapshot of feelings. A retention curve shows behaviour over time. Take everyone whose first order was in the same week, and plot the share of them who order again in week 1, week 2 and so on. Retention and cohorts shows how to build the table; here the question is what the shape tells you about fit.

• It drops, then flattens. Many people try and leave; that is normal. A flat tail means a core group found lasting value. This is the shape you want before you spend on growth.
• It keeps falling toward zero. No group found lasting value, or the product breaks its promise after a while. Growth spending here is wasted.
• Newer cohorts sit above older ones. The product is getting better for new users. Compare cohorts in the same chart.

Measure at the natural frequency. For the Mangrove app, count customers who placed at least one order in each week. For partner shops, count shops that ordered in the app in each fortnight. Daily active users would make a weekly grocery habit look broken.

How high should the flat tail be? Lenny Rachitsky collected benchmarks from many companies in 2020. At six months, he put good user retention at about 30% for consumer transactional products (where people buy something each time), about 25% for consumer social apps and about 60% for products sold to small and mid-sized businesses. Great was about 50%, 45% and 80% respectively. Use these as rough guides only. The level that matters for you is the one your unit economics can carry: if each lasting customer earns back what it cost to acquire them, a flat curve can fund growth (Unit economics).

Two traps distort the curve. First, promotions: a 50%-off first order fills week 0 with people who were never going to stay, so the curve looks worse than the product is, or, if the coupon repeats, better. Compare cohorts with and without the promotion. Second, mixing segments: a strong central-city segment averaged with a weak suburban one can look like a mild, slowly falling curve that describes nobody. Split by segment before you judge.

Fit when users cannot leave: internal tools and B2B apps

Store staff must use the Store app. Riders must use the Rider app. When use is mandatory, retention is 100% and tells you nothing, and a survey can be polite rather than true (Users, customers and stakeholders). You need other signals of fit, ones that show whether the tool has become the easiest way to do the job.

• Workarounds die. Count the work still done outside the tool: paper forms, photos in chat groups, side spreadsheets, phone calls. When the tool fits, these fall without anyone banning them. This is the internal version of retention.
• Use beyond the mandate. Staff use the tool for cases where the old way is still allowed, or for tasks it was not built for.
• Pull. Stores or teams outside the pilot ask for it before the rollout reaches them. Managers complain loudly when it is down.
• The outage test. When the tool fails for an hour, do people wait for it to come back, or go back to paper with relief?
• Time and errors. The task gets faster and needs fewer corrections, measured against a baseline (Product management for internal tools).
• An adapted survey. "If we removed the new receiving screen tomorrow and you went back to paper, how would you feel?" Ask it anonymously, of the people who do the work, not their managers.

Here is how that looked for a new scan-to-receive flow in the Store app, piloted in 12 stores for six weeks. Receiving a supplier delivery had taken about 51 minutes, 21 of them on workarounds. After six weeks the median was 39 minutes. Photos of delivery notes posted in the pilot stores' chat groups fell from about 30 a day to 4, with no rule against them. Goods received notes (GRNs) corrected later fell from about 1 in 12 to 1 in 30. Three managers outside the pilot asked to join. When sync failed for two hours one morning, staff queued the deliveries and waited rather than reach for paper. Of 70 receiving staff surveyed anonymously, 61% said they would be very disappointed to go back. That is fit, and it justified a rollout plan (Change management and adoption).

B2B apps sit in between. Partner shops can always phone a sales rep instead of using the Partner app, so they have a real choice, and the survey works. But a shop is not one person. The owner may love the credit view while the assistant who places orders still prefers the phone. Measure fit as the share of each shop's orders placed in the app, not the number of shops that installed it. Today about 62% of partner orders come through the app. A shop that places 9 of its 10 orders in the app has adopted it; a shop that tried it once has not.

When to scale: score fit, then decide

Before you spend to grow a segment, a city or a rollout, check four signals for that segment, and write down the threshold for each in advance:

1. Attitude: about 40% or more "very disappointed" among the right users, with at least 40 answers.
2. Behaviour: the retention curve flattens at the natural frequency, at a level your unit economics can carry. For internal tools, workarounds fall.
3. Pull: a rising share of new users arrives without paid effort, through friends, referrals or other teams asking for it.
4. The promise holds at scale: the thing the very disappointed users value still works at peak load. For a 30-minute delivery app, that means it still arrives in 30 minutes on a rainy Friday evening.

Then make one of three decisions, and date the next check:

• Scale: all four signals pass. Spend on growth in this segment, and keep measuring each new cohort, because new users may be less keen than early ones.
• Fix: people who use it value it, but a gap stops more of them. Keep growth spending low and work on the gaps the somewhat disappointed name.
• Narrow: one segment shows fit and others do not. Focus on the segment with fit, and stop spending on the rest until the product changes for them.

A weak score does not always mean the product is wrong. Sometimes people never reach the value, because sign-up is slow or the first order goes badly. That is an onboarding problem, and it is the subject of the next lesson, Onboarding and the first-time experience.