6. Retention before growth

§The question: do the people who reach the value come back, and does a group of them keep coming back forever?

§This is the most important chapter in the whitepaper, because retention is the assumption most first versions get wrong and the one that everything else depends on. A product people do not return to is not a product. It is a demo that charged for a first visit.

§ 6.1The retention curve#

§Take a cohort, the users who activated in a given week, and plot the share still active one week later, two weeks, four, eight, twelve. The shape of that curve is the single most honest picture of whether you have built something.

§Three shapes. A curve that decays to zero means no one found lasting value; the product is a leaky bucket. A curve that flattens at some level above zero, a "smile" if it ticks back up, means a group of users found value that persists; that flat portion is the fraction of people for whom the product became part of their life. A curve that flattens high is a great product.

§Andrew Chen's framing is the one to hold: the flattening, not the starting height, is what matters. A cohort that starts at 80% and decays to zero is worse than one that starts at 40% and flattens at 25%, because the second has 25% of users who will apparently never leave, and the first has none.

§ 6.2Why it comes before growth#

§Brian Balfour's essay put it plainly: retention is the foundation, because it drives everything downstream. It drives revenue, because retained users pay longer. It drives acquisition, because retained users refer. It drives the unit economics from the fifth whitepaper, because lifetime value is retention made into money. A team that improves retention improves every other metric at once. A team that improves acquisition while retention leaks improves nothing that lasts.

§This is why the whitepaper forbids growth spending until the retention curve flattens. Sean Ellis's precondition for growth is exactly this curve. Before it flattens, every dollar of acquisition buys a user who leaves, and the faster you acquire, the faster you confirm you have nothing.

§ 6.3Reading the curve honestly#

§The curve must be a cohort and it must be long enough. A four-week window on a product people use monthly shows a decay that is really just the usage interval; the window has to match how often the product is meant to be used. Lenny Rachitsky's benchmark work is useful here mostly as a caution: "good" retention varies enormously by category, and comparing your curve to a benchmark from a different usage frequency is how teams panic or relax for the wrong reasons. Compare your cohorts to your own earlier cohorts. Improvement over yourself is the signal.

§ 6.4When retention will not flatten#

§Sometimes the curve keeps decaying no matter what you try. This is the hardest finding in the whitepaper to accept and the most valuable, because it is the market telling you, in behavior, that the value is not there. The response is not another onboarding experiment. It is the iterate-or-pivot decision from the fourth whitepaper, made honestly, while there is runway. A team that keeps running retention experiments on a curve that will not flatten is refusing to hear the answer.

§ 6.5What you leave with#

§A retention curve for each cohort, over a window that matches the product's usage frequency, compared to earlier cohorts and to a control where one exists. An honest reading of whether it flattens. And a hard rule, agreed in advance: no growth spending until it does. The next chapter adds the second half of viability, whether people pay.