The MVP was the question. What happens in the three months after launch is the answer, and it is the period where most early companies make their most expensive mistake: reading weak signal as a feature problem and building their way further from a market.

Almost always, the right move at this stage is to narrow rather than to add.

The one number that means something

Retention. Not sign-ups, not traffic, not a chart that goes up because you launched.

The question is whether people who used the product come back and use it again, at the natural frequency of the problem you solve. A weekly tool should be used weekly. An annual tool cannot be judged on a month of data, and if your product has an annual cadence you need a different proxy entirely.

What a healthy curve looks like: usage drops after the first period — it always does — and then flattens. A flat section means a group of people found something durable. The height of the flattening matters less than its existence.

What an unhealthy curve looks like: a smooth decay toward zero with no flat section. No group has found durable value. Adding features to this curve does not bend it, and this is the single most common way early-stage money gets spent badly.

For rough calibration, benchmarks compiled by Lenny Rachitsky from a panel of operators put six-month user retention at around 25% for good consumer social and 45% for great; 30% and 50% for consumer transactional; 40% and 70% for consumer SaaS; 60% and 80% for SMB and mid-market SaaS; and 75% and 90% for enterprise SaaS. Net revenue retention over twelve months runs differently again — around 100% for good bottom-up SaaS and 120% for great.

Use these as orientation. Your own curve's shape matters more than its position against someone else's median.

The survey that puts a number on the feeling

Alongside behaviour, ask the question Sean Ellis devised and Rahul Vohra built a method around at Superhuman: how would you feel if you could no longer use this product? Very disappointed, somewhat disappointed, not disappointed.

Ellis found, across close to a hundred startups, that 40% answering "very disappointed" was the threshold separating products that could grow from products that struggled. Hiten Shah's 2015 study of Slack found 51%.

Two things make this more than a vanity metric.

Segment the answers. Vohra's most useful insight was that a headline score hides the real picture. Superhuman's overall score was 22%. When he segmented respondents and focused on a specific persona — the high-expectation customer — the score for that group was 32%. The product was not failing; it was succeeding with a narrower group than the company had been targeting.

Use the fence-sitters, ignore the rejectors. He discarded the "not disappointed" responses entirely and analysed only the "somewhat disappointed" users whose primary benefit already aligned with the product's core strength. Their complaints identified what was blocking them from becoming lovers. Everyone else's complaints would have led the roadmap somewhere unprofitable.

Superhuman's score moved from 22% to 58% over three quarters on a roadmap split roughly half toward deepening what enthusiasts already valued and half toward removing the obstacles the fence-sitters named.

Narrowing beats adding

The instinct on weak numbers is to build more. The evidence usually points the other way.

If ten users out of a hundred love the product, the question is not what would the other ninety need. It is what do those ten have in common, and are there ten thousand more of them.

Look for a shared characteristic you could actually target: a company size, a role, a workflow, a trigger event, a tool they already use. If you find one, the correct response is to rewrite the positioning around that group, change who you market to, and possibly remove features that only exist for the people who left.

This feels like shrinking the business. It is the opposite. A product that is essential to a definable group has somewhere to grow from. A product that is mildly useful to everyone has nowhere to stand.

If you cannot find a shared characteristic among your enthusiasts, that is a genuinely important finding and it points back toward the problem rather than the product.

Talk to three groups, and they are not equally useful

People who stayed. What did they replace, what would they miss most, when did it click. This produces your positioning language, and the words they use are better than anything you will write yourself.

People who signed up and stopped. The most valuable and the most avoided. They are easier to reach than founders expect — most will reply to a short, genuinely curious message from a founder. The answer is frequently mundane: they got confused, they forgot, the moment passed. Mundane answers are fixable.

People who never signed up. Hardest to reach, and they tell you about your message rather than your product.

Keep doing this at volume. The temptation is to treat customer conversations as a pre-launch activity. The teams that get to fit keep running them long after.

What to measure, and what to ignore

Worth watching:

  • Retention by weekly or monthly cohort, plotted as a curve rather than a single number.
  • Time from sign-up to first real value, and the percentage who never reach it.
  • The proportion of users doing the core action more than once.
  • Where in the funnel people drop, specifically.
  • Qualitative reasons attached to cancellations.

Worth ignoring at this stage:

  • Total registered users. Includes everyone who left.
  • Traffic, unless you are testing a channel deliberately.
  • Feature usage breadth. Users touching many features is not the same as users returning.
  • Anything that only ever goes up. Cumulative charts cannot show you a problem.

Give it a decision structure

Open-ended iteration drifts. A simple rhythm that works:

Monthly, look at the cohort curve and the survey score for your chosen segment. Ask one question: is the flat section higher than last month.

Quarterly, make an explicit call — continue, narrow, or change. Write down beforehand what result would trigger each. Deciding after seeing the data is how companies spend two years on a product that never had a flat section.

Split the roadmap. Vohra's half-and-half division is a good default: half toward deepening what the enthusiasts already love, half toward the specific obstacles named by users who nearly love it. Requests from people outside both groups go on a list and stay there.

When it is working

Todd Jackson's formulation is the most useful description we know: if you have to ask whether you have product-market fit, you do not. The reports from teams that reach it are consistent and slightly absurd — demand outruns the ability to serve it, customers arrive unprompted, the bottleneck moves from finding users to supplying them.

Before that point, the work is narrowing, sharpening the message, and removing whatever stops the right people from getting to value. It is less satisfying than building features and it is what actually moves the curve.

The message work in particular tends to be underestimated. Once you know which segment the product is for, the site, the onboarding and the positioning usually need rewriting around that group rather than the general audience they were written for — and that rewrite frequently improves conversion more than a quarter of engineering would have. That is most of what we do inside Grow: sharpening the message around who the product is demonstrably for, and fixing the path between arriving and understanding.

If you launched recently and the numbers are ambiguous, the first thing to do is not to build. It is to find out whether there is a flat section hiding inside your average, and who the people in it are.

Common questions

How do I measure product-market fit?

Two instruments together. Behaviourally, plot retention by cohort and look for a flat section — a group whose usage stops decaying. Attitudinally, ask users how they would feel if they could no longer use the product, and track the percentage answering "very disappointed" within your target segment.

What is the Sean Ellis 40% test?

A survey asking users how they would feel if they could no longer use the product. Ellis found across close to a hundred startups that 40% answering "very disappointed" separated products that could grow from those that struggled. Superhuman moved from 22% to 58% over three quarters using it as a roadmap input.

What does a good retention curve look like?

It drops after the first period — always — and then flattens. The flat section is the signal: a group has found durable value. A smooth decay toward zero with no flattening means no group has, and adding features will not bend that curve.

What retention rate is good for a SaaS product?

Operator-panel benchmarks put six-month user retention at roughly 60% for good SMB and mid-market SaaS and 80% for great, rising to 75% and 90% for enterprise. Consumer products sit considerably lower. Your curve's shape matters more than its position against someone else's median.

Should I add features or narrow my audience after launch?

Narrow first, almost always. If a minority of users love the product, find what they have in common and whether there are many more of them, then rewrite the positioning around that group. A product essential to a definable segment has somewhere to grow from; one mildly useful to everyone does not.

How long does it take to reach product-market fit?

There is no dependable timeline, and treating it as a milestone with a date tends to produce premature scaling. The more useful discipline is a quarterly decision point with the triggers for continue, narrow or change written down in advance.

Sources and further reading