The funnel article placed product-market fit at the leakiest stage of a startup, and the first article in this series placed its absence, no market need, at the top of the list of causes of death. This article asks what fit actually is, and answers that it is not a feeling and not a launch, but a measurable property of retention. A product has fit when a cohort of the people who try it keeps coming back, so that the retention curve flattens above zero instead of decaying to nothing. That single distinction, a curve that settles at a plateau against a curve that falls to zero, is the quantitative heart of the matter. The treatment is general, and it is information rather than business advice.

A Brief History

The phrase entered common use through the venture investor Marc Andreessen, who in 2007 called product-market fit the only thing that matters for a young company, crediting an earlier formulation by Andy Rachleff. The underlying counsel, to build something people want, is older than the phrase, and the lean-startup movement had already recast the search for fit as a process of validated learning rather than a single act of insight. Measurement came in stages. An early and still useful heuristic, proposed by Sean Ellis, asked existing users how they would feel if they could no longer use the product, and read a large share answering very disappointed as a sign of fit. The analytics era moved the test from surveys to behavior, to the cohort analysis of whether users actually return, which is the measure this article develops.

What Fit Actually Is

Fit is the demonstrated pull of enough of the right people wanting a product strongly enough to keep using it. Each phrase carries weight. Enough, because a handful of enthusiasts is not a market. The right people, because pull from users who will never pay or never recur does not sustain a company. And keep using it, because a product that is tried once and abandoned has no fit, however large the trial. Before fit, a founder pushes the product onto an indifferent world. After fit, the world pulls the product out of the founder faster than it can be supplied. The felt difference is real, but a feeling is not a measurement, and the measurement that matters is whether the people who arrive choose to stay.

Retention as the Signature

Consider a cohort, the group of users who first try a product in the same period. Let $R(t)$ be the fraction of that cohort still active $t$ periods later, so that $R(0) = 1$ because everyone is present at the start. The shape of $R(t)$ as $t$ grows decides the question of fit, and there are only two fates.

In the first, the curve decays toward zero,

\[R(t) = e^{-\lambda t} \longrightarrow 0,\]

an exponential decay in which every cohort empties out completely, given enough time. A company in this state is a leaky bucket, held at any level only by constant acquisition, since every cohort it contains eventually leaks away. In the second, the curve falls at first and then settles at a positive level, a plateau that does not decay. A stable core of users has made the product part of their routine. The flattening of the retention curve above zero is the signature of product-market fit, and its absence, a curve that slides to the floor, is the signature of its lack.

The Plateau Model

A simple model captures both fates at once. Suppose a fraction $f$ of each cohort becomes loyal and effectively never leaves, while the remaining fraction $1 - f$ churns away at a constant rate $\lambda$. The retention curve is then

\[R(t) = f + (1 - f)\, e^{-\lambda t}.\]

At the start $R(0) = f + (1 - f) = 1$, and as time grows the second term vanishes, leaving the plateau

\[\lim_{t \to \infty} R(t) = f.\]

Product-market fit, in this model, is simply the condition $f > 0$, the existence of a durable core, and the size of $f$ measures how much fit a product has. Take an illustration with a loyal fraction of $f = 0.3$ and a monthly churn of $\lambda = 0.5$ among the rest. The curve reads about $0.72$ after one month, $0.46$ after three, $0.33$ after six, and $0.30$ after a year, visibly settling toward thirty percent. A product without fit, $f = 0$, retains $e^{-6} \approx 0.002$ of its cohort after the same year, which is to say none of it. Thirty percent forever and zero percent are different businesses entirely.

The Leaky Bucket

Retention governs how large a business can grow. In the simplest account, a user base $N$ gains new users at a rate $a$ and loses a fraction $\lambda$ of its members each period, so that

\[\frac{dN}{dt} = a - \lambda N.\]

The base stops growing when arrivals balance departures, at the steady state

\[N^\ast = \frac{a}{\lambda}.\]

The churn rate $\lambda$ sits in the denominator, which is the whole point. At a monthly acquisition of a hundred users and a churn of one half, the base settles at two hundred, no matter how long the company runs. Lower the churn to one tenth, the improvement that fit provides, and the same acquisition settles at a thousand, five times the business from the very same spending. The point here is narrower and prior. Without retention, acquisition leaks away, and the business cannot fill.

Lifetime and the Value of a Plateau

Retention also sets the value of a user, through the time that user stays. The expected lifetime is the area under the retention curve,

\[L = \int_0^\infty R(t)\, dt,\]

a standard fact from survival analysis for any quantity that endures over time. For a cohort that only churns, $R(t) = e^{-\lambda t}$, the area is $1 / \lambda$, just two periods at a monthly churn of one half. A plateau transforms this. The flat plateau of the previous section is the idealized limit in which the loyal core never leaves and the lifetime grows without bound. Any real core churns slowly, at some small rate $\mu \ll \lambda$, so that

\[R(t) = f\, e^{-\mu t} + (1 - f)\, e^{-\lambda t}, \qquad L = \frac{f}{\mu} + \frac{1 - f}{\lambda}.\]

With a loyal fraction $f = 0.3$ churning at $\mu = 0.02$ each month, against a remainder at $\lambda = 0.5$, the lifetime is about $15 + 1.4 \approx 16$ months, some eight times the two months the same product earns without fit. Nearly all of that lifetime comes from the small loyal core, which is the value a plateau creates. A longer life is worth more revenue over its course, and the value side of that ledger, weighed against the cost of acquisition, belongs to the article on distribution and getting paid.

Measuring Fit

The honest measure of fit is the cohort retention curve, read out far enough in time to see whether it flattens or falls. The survey heuristic has its place as an early signal, before there is enough history to plot a curve, but it is a leading indicator and not the thing itself. The danger throughout is the vanity metric. A total of registered users rises even as every cohort churns out, because new arrivals hide the losses, and a founder who watches the total can mistake a leaking bucket for a full one. The cohort view defeats this illusion, because it follows a fixed group over time and cannot be flattered by fresh acquisition. Other honest signs accompany a flattening curve, unprompted word of mouth, organic growth that does not depend on paid reach, and a rising frequency of use, each of them a different shadow of the same retention.

Why Fit Gates Everything

The funnel explained why this stage gates the rest. A leak here discards everything spent downstream. Distribution carries users into a product that does not hold them, and monetization charges a base that does not persist, so that effort at the later stages returns almost nothing until the curve flattens. This is why the search for fit precedes the scaling of anything. It is also why fit, once found, is the beginning of durability, since a base that stays is a base that a rival must pry loose rather than merely attract, a thread the article on moats takes up at the end of the series. The product that earns this retention is built under deep uncertainty, which is the subject of the next article on execution risk.

Epistemic State

The retention models here are stylized. Real churn is not constant, since the chance of leaving usually falls the longer a user stays, which bends the curve beyond the simple exponential. The clean split into a loyal fraction and a churning remainder is an idealization of a continuum of loyalties, and the true plateau is rarely flat forever. What counts as enough retention depends on the category, since a daily social product and an annual tax product flatten at very different heights and over very different horizons. The survey threshold is a rule of thumb, not a law, and fit is a matter of degree rather than a line crossed once. What survives these cautions is the qualitative claim, which the particular numbers only illustrate. A retention curve that flattens above zero is fit, and one that decays to zero is its absence. Throughout, this is general information, and it is not business advice.

Out of Scope

The unit economics that weigh the lifetime value of a retained customer against the cost to acquire one belong to a later article. The engineering and product work of building toward fit belong to the next article. The specific tooling of cohort analytics, the category benchmarks for what retention is good, and the mechanics of growth loops are practical subjects left to the operational literature and not treated here.

Conclusion

Product-market fit is not a mood and not a milestone announced at launch. It is a measurable property of retention, a cohort curve that flattens above zero because a durable core of the right people has chosen to stay. The plateau is the thing to find, its height is the degree of fit, and its presence is what makes every later stage worth attempting. Without it, acquisition leaks, monetization charges a vanishing base, and the funnel ends here, where the record says most ventures are in fact lost. The next article turns to the execution risk of building the product that earns the curve.

References