10. From funnel to loop

§The question: where does the current product leak, and how could its own output feed its next users?

§The first whitepaper introduced these two shapes; this chapter runs them. The validation signals held, growth is justified, and the first job is to see the business as a system rather than a series of features.

§ 10.1The funnel diagnoses#

§Dave McClure's pirate funnel, acquisition, activation, retention, revenue, referral, is a diagnostic. Each stage asks a question, and each stage keeps a fraction of the people who entered it.

§Acquisition: do people arrive? Activation: do arrivals reach first value? Retention: do they come back? Revenue: do they pay? Referral: do they bring others? A cohort enters at the top and shrinks at each stage, and the shape of the shrinkage tells you where the business is broken. A funnel that loses everyone at retention is a retention problem no acquisition can fix, which is the whole argument of Part II restated as a diagram.

§Use the funnel first, because it points precisely at the leak. Growth experiments applied anywhere but the leak are wasted, and the funnel is how you find the leak.

§ 10.2The loop compounds#

§A funnel is a leak by nature: you pour in at the top, and to grow you must pour more, forever. A loop is different in kind. In a loop, the output of one turn becomes the input of the next: users produce something, content, invitations, data, reputation, that brings or activates more users, who produce more of the same.

§Reforge's argument that loops are the new funnels is the framing to hold. A funnel's growth is bounded by spend. A loop's growth compounds, because each cohort makes the next cohort cheaper or larger. The loops the first whitepaper named are the engines: a content loop where users create pages that rank and attract users who create more pages; a viral loop where users invite users; a paid loop where revenue funds acquisition that produces more revenue.

§ 10.3Funnel to find the leak, loop to build the engine#

§The two are not rivals; they are used in sequence. The funnel diagnoses which stage is weakest, and the growth experiments in the next five chapters fix that stage. But the goal of all of them is to close a loop, so that improvement at one stage feeds the others.

§The clearest AI-era examples are loops. A tool where every generated artifact is shareable and every share is an ad for the tool. A product where every user's data makes the model better for the next user, so retention improves acquisition through quality. An assistant that produces public answers that rank in search and bring the next user. Each is a loop where the product's use is its own distribution.

§ 10.4Why loops matter more in the AI era#

§The cost structure argument from the fifth whitepaper returns here. When serving a user has a real marginal cost, inference, a funnel that requires ever more paid acquisition against thin margins does not survive contact with the spreadsheet. A loop that grows without proportional spend is not just better; for many AI products it is the only growth that the unit economics permit. The teams that endure are the ones whose product's use is its own distribution.

§ 10.5What you leave with#

§The current funnel, by cohort, with the weakest stage identified as the target for the next experiments. And a hypothesis about which loop the product could close, so that the stage experiments are aimed not just at plugging a leak but at building an engine. The next chapters take the funnel stage by stage.