Growth loops explained, with the four loop types and a template to draw your own

A growth loop is a system where an input, such as a new user, takes an action that produces an output, and that output becomes the next input, so each cohort of users brings in the next cohort. Reforge sorts acquisition loops into four types: viral, content, paid and sales. As an example, not a benchmark, 1,000 signups that bring in 100 more give a yield of 0.1 per cycle.

What is a growth loop?

A growth loop is an input, an action and an output, where the output becomes the next input. In plain terms, each cohort of users brings in the next cohort. Reforge describes 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." (Reforge)

The term comes from a Reforge series co-written by Brian Balfour, Casey Winters, Kevin Kwok, and Andrew Chen. The authors offered loops as the alternative to the AARRR (Acquisition, Activation, Retention, Referral, Revenue) funnel, which they note was originally created by Dave McClure.

Why the funnel was not enough

Their core objection is direction.

A loop changes the question your team works on. Instead of asking how to fill the top, it asks: "How does one cohort of users lead to another cohort of users?" (Reforge)

There is a practical caveat. The Growth Mind newsletter says that in theory a loop with output greater than input can "feed itself organically", but that in practice feeding it continuously remains essential.

The three parts, using one real loop

Reforge's Pinterest example maps onto the input, action and output structure.

Part	What it means	Pinterest
Input	A user enters or returns	User signs up or returns
Action	The user does something that creates value	Saves new content or repins existing content
Output	That value reaches new people	Pinterest distributes content to search engines, and a new user finds it and signs up

A worked example

Take round numbers as an example, not a benchmark. If 1000 users sign up this month and their activity brings in 100 signups next month, each user brings in 0.1 new users (100 divided by 1000).

In our reading, that loop adds to growth but cannot carry it alone. You still need other input, such as paid or outbound, to keep volume up while you work on the action step.

A starting template

Before you draw your loop, answer the three fundamental questions The Growth Mind sets out:

1. What action triggers the loop. The action reflects receiving, creating or distributing value.
2. Who performs it. Usually a user, sometimes a supplier, partner or someone inside the company.
3. Why they do it. The motivation behind the action: financial, social or personal.

Write one sentence for each, then turn every weak point into a test in your growth experiment backlog. If the weak point is the action step, check your activation rate first.

How is a growth loop different from a funnel?

The earlier section showed that a loop feeds its output back in as input. A funnel stops at the bottom. In our reading, the funnel most teams draw is the AARRR framework (Acquisition, Activation, Retention, Referral, Revenue).

Question	Funnel (our summary)	Loop (our summary)
Which way does it run?	Top to bottom, then it ends	Output returns as the next input
How do you grow it?	Put more in at the top	Reinvest what comes out
Who owns it?	One team per stage	Product, marketing and sales together
What is it good for?	Diagnosing one step	Explaining how the product grows

The reinvestment row is concrete in a paid loop. In Conor Dewey's summary of the Reforge course, revenue from paying users is reinvested in more ads. A funnel has no step for that, which is why, in his words, funnels push us to invest in linear activities.

The silos funnel thinking creates

Reforge names two. The first is strategic. Product and acquisition get planned apart: the product team goes off and plans their product strategy and then marketing goes off and creates the acquisition strategy. Reforge also says teams treat our monetization strategy in a third silo.

The second is functional. Teams get structured by funnel layer: Marketing owns acquisition. Product owns retention. Sales (if B2B) owns revenue. (B2B means business to business.) Each team chases its own metric. Reforge describes the result: marketing brings in low quality users/leads at top of funnel to hit their goal, but that tanks retention.

If your team goals map one to one onto funnel stages, we think it is worth adding one shared loop metric this quarter, such as new users produced per cohort of existing users.

When the funnel is still the right view

Keep the funnel for a single step. Reforge says the funnel helps explain a specific step within a Growth Loop, but misses the larger picture of the loop itself.

Here is a worked example with round numbers, not a benchmark. In a content loop, if 1000 people view a shared page and 100 sign up, that step converts 10 percent. The funnel tells you where that step leaks. The loop tells you whether fixing it brings more content, more viewers and more signups next cycle.

This week, draw your loop, pick the weakest step, and run a funnel analysis on that step alone. Check it against activation rate benchmarks, and size any test with the A/B test sample size calculator before you ship it.

What are the four types of growth loops?

Reforge sorts acquisition loops into four categories: viral, content, sales and paid. Most companies do not run all four at full strength. In the same summary, one or two loops usually carry most of the acquisition load, so the first job is to find yours.

Loop	Input	Action (user or company)	Output	Step that usually limits it (our reading)
Viral	New users	User invites or shares through a channel	Invitees who respond and sign up	Response rate on the invite
Content	New users or company resources	Users, suppliers, partners or the company create and distribute content	Viewers who sign up	Distribution, by search or social
Paid	New users	Some users convert to paying	Ad budget that brings in new users	Conversion to paid
Sales	New paying customers	Company uses profits to hire more sales reps who work leads	Leads that convert to customers	Rep ramp time and lead supply

For content loops, Reforge splits the work two ways: who creates it (user-generated or company-generated) and who distributes it (SEO, meaning search engine optimization, or social).

What drives a viral loop

Viral loops run on motive. Reforge describes users who share because they want to gain personal, financial, or social capital:

• Personal. The product gets better for the inviter when their network joins. Personal capital includes entertainment, communication, information, and time saved.
• Financial. Users share for monetary or financial benefits, the logic behind a referral program.
• Social. Users share for social capital or community engagement.

Reforge measures loop efficiency with the K-factor, which is the number of invites sent per new user multiplied by the percentage of the invites that convert. Andrew Chen gives a worked case: 1000 active users invite 500 coworkers, a viral factor of 0.5. Those 500 then invite 250, then 125. His multiplier formula is 1/(1-v), so 1/(1-0.5) gives a multiplier of 2.

Why viral growth depends on retention

Chen argues that virality is inextricably linked to retention. A user who comes back almost daily gives you 30 shots a month to prompt a share or invite. A one-and-done user gives you one.

This week, do two things. Write down which motive your invite flow relies on. Then check your retention against monthly churn benchmarks before you rework the invite screen. We think a weak return rate caps a viral loop sooner than a weak invite flow does.

Growth loop examples from companies that documented them

Each one maps to a loop type from the previous section, so you can see the trigger, the actor and the output in a real product.

Company	Loop type	What restarts the cycle (our reading)
Grubhub	Content, SEO distribution	Restaurant pages that rank for local cuisine searches
Pinterest	Content, SEO distribution	User boards that search visitors find, then add to
Substack	Content, user-generated	Writers whose work draws in other writers
Twitch	Content, creator-driven	Streamers who attract viewers

Grubhub and Pinterest: content loops built on search

When Casey Winters joined Grubhub, it had 30,000 users. By the time he left, it had three million users. The loop started with landing pages that organized restaurants into regions and types of cuisine.

At Pinterest, the trigger was a lost channel. Facebook removed the default ‘auto-share’ feature, and Pinterest had to replace it. The company had 40 million active users, but growth was tapering off.

Here is a check you can run this week. Split signups by traffic source. As an example, not a benchmark: if 1000 search visitors produce 100 signups, that source converts at 10%. Compare that rate with your other sources. A gap points to an onboarding mismatch like Pinterest's.

Substack and Twitch: creators feed the loop

Substack enables writers to easily share and monetize their work. As they do, it gets the attention of other writers, who discover Substack, join, and continue the loop.

On Twitch, streamers drive growth by attracting viewers. Those viewers join the platform, generate revenue, and watch other streams. The weak point sits on the supply side. Rivals compete with them for streamers, often with large exclusive contracts. We think any creator loop carries this risk, because the creators are the input.

LinkedIn: an AI-assisted content loop

Brian Balfour reported that LinkedIn's AI-assisted loop went from generating 0 to 1M uniques per month according to Ahrefs between March and June. In our reading, this is the same content and SEO pattern as Pinterest, with AI assistance in how the content gets made.

To draw your own version, write each case's trigger, actor and output into the growth experiment backlog template, then list one test per step.

How do I measure a growth loop's cycle time and yield?

Two numbers describe any loop. Cycle time is the days from one input to the next input it produces. Yield is the new inputs each input creates per cycle. A short cycle with a low yield can still compound, and a high yield with a slow cycle can feel stalled for a quarter.

Define the two numbers on one cohort

Take a cohort of new users who all entered in the same week. That week's signups are your input.

Cycle time is the median number of days between a user's own signup and the first signup they bring in. Use the median, not the average, because a few slow referrers will stretch an average.

Yield is the new inputs the cohort produced in one cycle, divided by the cohort size. As an example, not a benchmark: a cohort of 1000 users that brings in 100 new signups in its first cycle has a yield of 0.1, because 100 divided by 1000 is 0.1.

Why retention changes both numbers

A user who churns after one session gets one chance to produce an input. A user who keeps coming back gets many. Andrew Chen puts it this way: each visit is a small opportunity to prompt a share or an invite, and a sticky product can take its time, building viral growth over weeks and months.

In our reading, that means yield should be measured per cycle across several cycles, not once at signup. If the cohort is still active in its second and third cycle, it keeps adding inputs. Track cohort retention next to yield, and compare it against monthly churn benchmarks.

Which events to track by loop type

PostHog lists what to measure for each loop. The column on the right is our reading of where those events sit in the cycle.

Loop type	What PostHog says to measure	Where it fits in the cycle (our reading)
Viral	signing up, getting started, and invites or referrals, plus the referral source of signups	Input, activation, output
Paid	the margin between the cost of customer acquisition (CAC) and how much revenue we receive from them (LTV)	Revenue that funds the next spend
Content	traffic, conversion, and usage events	Discovery, signup, new content created
Sales	trial usage, lead conversion, usage after conversion, and account-level analytics	Lead in, closed account, revenue for hiring

A six-step setup for this week

1. Name the input event, such as a signup or a published page.
2. Name the output event that creates the next input, such as an invite accepted.
3. Store the source user or source page on every new input so you can attribute it.
4. Compute median cycle time per weekly cohort.
5. Compute yield per cycle for at least three cycles.
6. Log any change you test in your growth experiment backlog and check activation against activation rate benchmarks.

A worked example of finding the step that limits a loop

Here is an invite loop for a business-to-business (B2B) collaboration tool. The inputs are our own example numbers, not a benchmark, so swap in yours as you read.

The loop in one cycle

Start with 1,000 signups. At a 40% activation rate, 1,000 signups give you 400 activated users. If 25% of those 400 send invites, you have 100 inviters. 100 inviters sending 3 invites each produce 300 invites. With 30% of 300 invites accepted, 90 new signups come back into the top of the loop.

Divide the 90 returning signups by the 1,000 you started with and the yield is 0.09 per cycle. The cycle time is 14 days.

Play it forward

Brian Balfour describes the habit this way: play the loop forward multiple cycles, then work backwards to what this cycle needs. At a yield of 0.09, the 90 second-cycle signups produce about 8 more, and those 8 produce less than 1. Three cycles at 14 days each take 42 days, and the loop adds under 100 users to your original 1,000. That is a small boost to a channel, not a growth engine.

Every step multiplies the others, so a proportional lift anywhere moves the yield by the same factor. Doubling any single step takes 0.09 to 0.18.

Step	Current value	Doubled value	New yield
Activation	40%	80%	0.18
Share who invite	25%	50%	0.18
Invites per inviter	3	6	0.18
Acceptance	30%	60%	0.18

Because the math is equal, the choice comes down to headroom and cost. In our reading, the share who invite is often the cheapest to move first: going from 25% to 50% means asking one more activated user in four to share, through an invite prompt at the moment a teammate is needed. Doubling activation to 80% is far harder. Check yours against our activation rate benchmarks before betting on it.

At 0.18, the first 1,000 signups return 180, which return about 32, which return about 6.

When the loop runs in reverse

Balfour warns that self-reinforcing systems can reverse on you and produce compounding destruction instead. In an invite loop, that looks like activation slipping. If activation falls from 40% to 30%, the yield of 0.09 drops by a quarter to 0.0675, and every later cycle shrinks with it.

This week, write your four step rates in the table above, pick the one with the most headroom and the cheapest test, and add it to your growth experiment backlog. Size the test with the sample size calculator. If the invite step wins, the referral program playbook covers the mechanics.

A template to draw your own growth loop

The structure is input, action, output: the action turns an input into an output, and the output becomes the next cycle's input. Before you fill the first row, write down what the triggering action is. The Growth Mind puts it plainly: this action must reflect the reception, creation, or distribution of value.

The blank template

Copy it into a document and fill one row per step. Step 1's input is the previous cycle's output.

Step	Input	Action (who does what)	Output	Metric	Current rate	Cycle time	Owner	Measured? (yes/no)
1
2
3
4
5

The "who" is often a user. It is not always one. In some loops it could also be a supplier, a partner, or even someone internal to the company who triggers the action. Name that person or team in the owner column, even when the actor sits outside your company.

One filled row for each loop type

The rows below are our own examples of a single step. Your loop will have three to five steps. Leave the rate and cycle time blank until you have measured them.

Loop type	Input	Action	Output	Metric	Current rate	Cycle time	Owner (our example)
Viral	New active user	Sends an invite	Invited signups	Invites accepted per user	Your measured rate	Days from signup to first invite	Product
Content	New user	Publishes a page that gets indexed	Search visitors who sign up	Signups per published page	Your measured rate	Days from publish to first signup	Content or SEO lead
Paid	Ad spend	Acquires a paying customer	Revenue to reinvest	Revenue recovered per dollar spent	Your measured rate	Payback period	Growth marketer with finance
Sales	Trial account	Converts to a paid plan	Revenue that funds more sales capacity	Trial to paid conversion	Your measured rate	Days from trial start to close	Sales lead

For the paid row, the cycle time is the Payback Period (the time required to recoup the investment made to acquire a user). Shorter payback means the cash comes back sooner for the next round of spend.

How to fill the rate column

Use counts, not impressions. Here is an example, not a benchmark: if 1000 visitors reach a page in a month and 100 sign up, the step's rate is 100 / 1000, or 10%. Write the counts next to the rate so the next person can check it.

Fill the template this week in four steps:

1. Draw the loop on paper and number the steps.
2. Pull last month's counts for each step's input and output.
3. Put one name in every owner cell. A step owned by two teams usually has no owner.
4. Turn the step with the lowest rate into tests in the growth experiment backlog template. For the activation step, compare against activation rate benchmarks.

How to find your company's growth loop this week

You can run this in one working week. You need five colleagues, a whiteboard or shared document, and the loop template from the previous section.

Step 1: Ask five people the same question

Reforge asks participants in its programs to go around their company and ask five different people to whiteboard one question: "How does your product grow?" Choose people from different functions, such as a founder, an engineer, someone in sales or support, whoever runs marketing and a recent hire. Each person draws alone, before anyone compares notes.

Expect a mess. Reforge reports that everyone has a different answer, or that an answer "represents only one piece of the puzzle," or that it "talks about the output of $$$, but not the inputs of usage."

Step 2: Reconcile the boards into one loop

Put the five drawings side by side and work through what you see.

What you see on the boards	What to do next (our reading)
Different answers	Keep the steps that appear on most boards. Mark the disputed steps for a data check.
Only one piece of the loop	Ask that person what happens before and after their step until the drawing returns to a new or returning user.
Revenue, with no usage	Rewrite each step as an action a user, supplier or partner takes that you can count.

Test the merged drawing against the question loops force you to answer: "How does one cohort of users lead to another cohort of users?" If your drawing cannot answer it, we think you have drawn a funnel.

Step 3: Decide which stage you are in

Brian Balfour describes a sequence most products go through. They begin by doing things that don't scale to get their first customers. Then they build growth loops that hold up over time.

Do not take the breakout cases as your model. ChatGPT reached 100M users in just 2 months, and Balfour called it "an interesting case that I think nobody should try to follow." In his view, founders often believe they have caught lightning in a bottle too, "But 99 out of 100 times that’s not true."

Our test: if most new users still arrive through work your team does by hand for each one, you are in the first stage. In that case, write down the manual step that brings in the most users and keep doing it. Your loop design should then aim to replace that step.

Step 4: Pick one constraint and test it

Choose the single step in the merged loop with the weakest conversion, using the measures from earlier sections. Balfour argues that resetting a loop requires cutting initiatives to "focus the fire power in a concentrated area with higher return potential." Apply the same idea at a smaller scale: one step and one experiment this week.

Write that experiment into your growth experiment backlog. If the weak step is the first meaningful use of the product, compare it against activation rate benchmarks before you set a target.

What this page does not cover

This page does not cover how to set up event tracking in a specific analytics product, or channel tactics for running paid, SEO or sales programs in detail. The figures in the worked examples are illustrations, not benchmarks. Nothing here is financial advice, including the parts on ad spend, payback period and reinvesting revenue.

Related on Race to Scale: A/B test sample size and duration calculator, with the formula; What is a good activation rate? Benchmarks by product type and how to define yours; Free-to-paid conversion rate benchmarks for freemium, opt-in trials and card-required trials.