What is a growth loop and how do you build one?
A growth loop is a compounding system where the output of one cycle becomes the input for the next — users beget users, revenue funds acquisition, or product usage generates content that attracts more users. Unlike a funnel, which is linear and exhaustible, a loop compounds over time. Building one starts not with engineering scale but with engineering delight at a tiny, deliberately constrained scale.
The core structure of a growth loop
Every growth loop has three components: a trigger (something that brings a new user into contact with your product), a value delivery moment (where the user gets enough value to take the next action), and an output (that action producing a new trigger for another user). The loop closes when the output feeds back into the top. Examples: Dropbox users invite friends to get more storage — that referral is the output that becomes the trigger. Airbnb hosts earn money, use that money, and word spreads — social proof and earnings are the output. The key question is: what does a satisfied user naturally do next, and can you route that behavior to produce a new user?
Most founders try to engineer growth before they've identified what the loop even is. They spend on paid acquisition or chase partnerships before they understand what a happy customer actually does after they're happy. Paul Graham's observation about big-company partnerships is instructive here — founders assume a deal with a large company will ignite growth, but six months later nothing has moved. Partnerships don't close loops; they delay the work of finding them.
To map your loop, start with your best current users and trace backwards and forwards. How did they hear about you? What made them stay? What, if anything, did they do that exposed another person to your product? That path — however messy — is your proto-loop. Your job is to reinforce the strongest steps and remove friction from the weakest ones.
Why you must start unscalably small to find the loop
The counterintuitive truth about growth loops is that you can't discover them at scale — you discover them by watching individual users so closely that patterns become unavoidable. Paul Graham's point about doing things that don't scale isn't just about hustle; it's epistemological. When you personally recruit, onboard, and support your first 20 users, you learn *why* they stay, *what* they tell their friends, and *which* moment of the product creates the 'aha.' That's the loop signal buried in the noise.
Facebook's early constraint to Harvard — and then to specific colleges — is a textbook example of this. By limiting supply and matching it precisely to a community, they achieved the density of engagement needed to see what a social loop looks like in action before expanding it. A social product with sparse, unconnected users shows you almost nothing about virality. The same product with a dense, interconnected cohort shows you everything.
The practical implication: resist the urge to broaden your target user until you can watch a loop close at least three times in a row in a small population. If you can't see the mechanism at 50 users, you won't find it by going to 5,000. Narrow your market to the point where you can observe cause and effect directly.
The four types of growth loops and which to build first
Viral loops trigger when users invite or expose others — the output is a referral or impression. These have the highest ceiling but require a product with inherent social value or a strong incentive to share. Paid loops use revenue to buy acquisition, then convert and monetize those users to fund more acquisition — the loop closes when LTV exceeds CAC with enough margin to reinvest. Content loops occur when user activity generates discoverable content (reviews, posts, portfolios) that attracts organic search traffic and converts it into new users. Product-led loops happen when using the product itself creates a forcing function for others to adopt it — think invoicing software where the recipient of an invoice becomes a potential user.
For most early-stage startups, the honest answer is that you don't get to choose your loop type — your product's core value proposition chooses it for you. A marketplace almost always needs a liquidity loop before any other loop is relevant. A B2B tool with a collaboration feature can exploit product-led loops. A media product lives or dies by content loops. Trying to force a viral loop onto a product with no natural social component wastes months.
Start by identifying which loop is latent in what you've already built. Interview users who came from referrals and ask what motivated the referral. Look at your organic search traffic and what pages drive it. Check whether your paid CAC is already lower than LTV — if so, you have a paid loop that just needs more capital to turn. Fix the loop that already exists before designing a new one.
The moment a loop becomes self-sustaining — and how to get there
Patrick Collison described the inflection point at Stripe as a shift from pushing a boulder to riding a train car with its own momentum. That transition doesn't happen through a single strategic decision — it happens when the output-to-input conversion rate of your loop exceeds 1. Technically, if each user cycle produces more than one new user (a viral coefficient above 1.0 for viral loops, or a reinvestment ratio above 1.0 for paid loops), the loop is self-sustaining. Below that threshold, the loop decays without external input.
Getting to self-sustaining requires attacking the loop's weakest conversion step. Most founders optimize the trigger (acquisition) because it's visible and measurable. But more often the bottleneck is the value delivery moment — users aren't reaching the point of delight that makes them take the output action. This is where the early, unscalable work pays off: the obsessive attention you gave to your first users to make them ecstatic also teaches you exactly which product moments drive the output action. Once you know that moment, you can engineer the product to deliver it faster and more reliably for every subsequent user.
The cultural dimension matters too. Paul Graham notes that when a startup genuinely commits to delighting users early, that orientation permeates the company's culture and makes scaling customer experience more tractable than founders expect. A growth loop built on top of genuine user love is structurally more durable than one built on incentives or friction — because users who are actually happy close the loop voluntarily, without being bribed.
How to instrument and iterate your loop
You can't improve what you can't see. Map each stage of your loop to a measurable event: trigger → acquisition event, value delivery → activation event, output → referral or revenue or content event. Then measure the conversion rate between each stage. The stage with the lowest conversion rate is your constraint, and that's where you direct all product and growth work until it improves.
One trap founders fall into is measuring aggregate metrics (total signups, total revenue) rather than cohort-level loop metrics. Aggregate numbers obscure whether the loop is closing. Instead, track what percentage of users acquired in week N produce at least one new user by week N+4. If that percentage is rising, your loop is strengthening. If it's flat, the loop isn't closing despite whatever growth you're seeing from paid or manual acquisition.
Iterate in short cycles — two weeks maximum — with a single change per cycle. The reason to test one thing at a time is that loop mechanics are interdependent: a change to onboarding affects activation rates, which affects referral rates, which affects acquisition. If you change three things simultaneously, you can't attribute the change in loop performance to any single decision. Slow, methodical iteration on one step at a time compounds faster in practice than rapid simultaneous experiments, because you accumulate real knowledge rather than correlated noise.
“It tipped from being this boulder we had to push to being a train car that in fact had its own momentum.”
— Patrick Collison, quoted by Paul Graham, source
The one thing to do
Identify the one action your happiest users take that brings in another user, remove every friction point blocking that action, and measure whether the conversion rate is rising — that is the only growth loop work that matters right now.
Frequently asked questions
How is a growth loop different from a marketing funnel?
A funnel is linear — you pour users in at the top and they exit at the bottom. A growth loop is circular — users who exit at the bottom generate new users who enter at the top. Funnels require constant external input to produce output; loops compound on their own once the conversion rate between stages exceeds 1.
When should a startup start thinking about growth loops?
As soon as you have any users at all — not to engineer the loop, but to observe whether one is forming naturally. The signal that a loop exists is when you find users who arrived because of another user's action rather than your direct outreach. Once you see that signal, you invest in reinforcing it.
What's the most common reason growth loops fail to close?
The value delivery moment is too slow or too weak — users don't reach genuine delight before they churn, so they never take the output action that would close the loop. Fixing this almost always requires doing things that don't scale: hands-on onboarding, direct support, and watching users interact with the product in real time.
Can a startup have more than one growth loop?
Yes, and the strongest companies usually do — a product-led loop running in parallel with a content loop, for example. But running multiple loops simultaneously before any single loop is self-sustaining is a focus trap. Close one loop first, then layer a second.
Sources
- Do Things that Don't Scale — Paul Graham
- How to Raise Money — Paul Graham