Growth Loops: A Builder's Guide to Compounding Growth

A growth loop is a closed system in which an input, usually new users, takes an action in your product that creates an output, such as an invite, shared content, or revenue. That output brings in or re-engages more users, who become the next input. Each cycle feeds the next, so growth compounds over time.

Every loop has four parts:

• Input: The users, content, or dollars entering the loop.
• Action: What those users do in the product that creates value, like sharing a file or answering a question.
• Output: The by-product of that action that reaches new people or funds new acquisition.
• Reinvestment: The step where output turns back into input.

Brian Balfour, founder and CEO of Reforge and former VP of Growth at HubSpot, popularized the framing in his essay Growth Loops are the New Funnels. He defines loops this way:

"Loops are closed systems where the inputs through some process generates more of an output that can be reinvested in the input."

The word that matters for builders is reinvested. If the output doesn't flow back in on its own, you've shipped a campaign, not a loop.

Growth loops compound because each cycle's output feeds the next cycle, while funnels move users in one direction and lose some at every step. In Balfour's framing, the customers who make it to the bottom are the end of the story, and growing means pouring more in at the top.

A loop treats those converted customers as the start of the next cycle. Their invites, content, or revenue become fresh input. That's why a loop can get cheaper per user as it grows, while a funnel tends to get more expensive.

The research backs the mechanism. Hill, Provost, and Volinsky found in a 2006 Statistical Science study that "network neighbors" of existing customers adopted a telecom service at three to five times the rate of groups chosen by the firm's own marketing playbook. It's one service from one company, so treat it as evidence that adoption travels along networks, not as a benchmark for your loop.

Funnels don't disappear. Every step inside a loop has its own mini funnel, and you'll use funnel analysis to see where users drop out of each one.

The main types of growth loops are viral, content, paid, and sales loops, and each one turns a different output back into new users. Pick the one whose output your users already create, and get it running before you add a second.

Viral loops

Users bring in other users by inviting them or by using the product with them. Andrew Chen, a general partner at Andreessen Horowitz who previously worked on growth at Uber, describes in The Atomic Network how Slack spread team by team inside companies, with a network of fewer than 10 people enough to get going.

Chen also flags the limit. For products that need a company-wide rollout before they deliver value, he writes:

"Here, a viral growth strategy is difficult, because it's not enough to dribble in a few users at a time."

Content loops

Users or the product create content, search and social distribute it, and new visitors arrive and create more. Balfour and Fareed Mosavat break down LinkedIn's AI-assisted growth loop, where AI lowered the effort to contribute content. They report the pages went from zero to 1 million unique visitors a month between March and June 2023, according to Ahrefs data.

Paid loops

Revenue from new customers funds ads that buy more customers. Picture a hypothetical subscription fitness app that reinvests first-month revenue into paid social. The loop holds only as long as you can float the ad spend until each new customer pays it back.

Sales loops

Product usage produces signals that point sales at the right accounts, and closed deals add seats that create more usage. In a hypothetical B2B project tool, any workspace that adds a third team gets flagged, sales reaches out, and the expanded contract brings in new users who repeat the pattern.

You build a growth loop by finding the loop your product already half-runs, defining a metric for each step, and shipping the reinvestment step as code you can measure.

Find the loop already in your product

Start with what users already do that touches non-users: shared links, exports, public pages, invites, embeds. Look at what happens right after the aha moment, the point product-led growth teams design around, when users are most likely to bring someone else in. Pick the output that already reaches new people, even if it's weak, rather than inventing a loop from scratch.

Define the input, action, and output metrics

Give each step one tracked event and one owner. For a hypothetical design tool, the input is workspace_created, the action is file_shared, and the output is invitee_signed_up. Record a baseline for each before you change anything.

Build and instrument the mechanism

Ship the reinvestment step as a product feature: the invite flow, the share-to-public toggle, the auto-generated landing page. Then attach properties like referrer_id and loop_source to every new user, so you can tell loop-acquired users from everyone else. Treat each release as one lap of the build, measure, learn loop.

To nudge users at the right step, Amplitude Guides & Surveys can trigger an in-product prompt based on behavior, for example only for users who shared a file but haven't invited a teammate. You can then see how users interacted with the prompt alongside your loop metrics.

You measure a growth loop by tracking conversion at every step, how much new input each cycle produces, how long a cycle takes, and whether loop-acquired users stay. Five metrics cover it:

• Step conversion: The share of users who move from each step to the next.
• Loop yield: New inputs generated per input. Above one, the loop sustains itself. Below one, it amplifies your other channels but still needs fuel.
• Cycle time: Median days from an input entering to its output arriving as a new input.
• Loop velocity: How many laps the loop completes in a period (days in the period divided by median cycle time).
• Loop cohort retention: How users acquired through the loop retain compared with users from other channels.

Cutting cycle time can matter as much as raising yield, because a faster loop compounds more times a year. And watch retention closely. A loop that brings in users who churn is a leak with extra steps, and retention curves split by loop_source show it fast.

In Amplitude, you can build each loop step as a funnel analysis chart. When a step stalls, Session Replay takes you from the drop-off straight into real sessions, with rage clicks and errors in context. Then test the fix with an experiment so you know the lift came from your change.

AI agents automate a growth loop by running sense, decide, act, and learn on its metrics, so a stalled step gets attention without waiting for the next planning cycle. The agent watches step conversion and cycle time, picks the weakest step and segment, ships a test or prompt, then reads the result before the next round.

That turns a loop you tune once a quarter into one that tunes itself, the core of a self-improving product. Balfour and Mosavat say they have mixed feelings about LinkedIn's AI-assisted loop because much of the content was thin. Automation scales whatever you point it at, so set the quality bar before you set the agent loose.

Amplitude AI Agents can run automated diagnostics that show what changed in your loop and why, including anomalies. They can also segment users, build cohorts for the stalled step, and bulk-analyze session replays to show what's behind it. Amplitude Wave, the proactive product agent, goes a step further: it analyzes your product data and code, surfaces what to build, and orchestrates plans, pull requests, and experiments, then measures the impact. Wave is in closed beta and rolling out to customers in the second half of 2026, so expect its scope to change and keep a human reviewing what ships.

Start with one loop, one owner, and events on every step. Once you can see yield, cycle time, and cohort retention, hand the repetitive watching to agents and put your time into the next loop.