Growth Loops

Everyone in a growth meeting has seen the funnel chart: visit, signup, activation, payment. Every month you have to put new people at the top of that funnel, and doing so takes either ad budget or constant effort from the team. Stop, and growth stops.

A loop runs on a different logic. A user uses the product, that use produces an output (a piece of content, an invite, some data, revenue), the output brings a new user, and that user does the same. Because the output returns as input, growth feeds itself. Designing it is closer to mechanism engineering than to planning a campaign.

When you need it

If activation and retention are solid and your growth is still linear (every new user requires extra spend or effort), it is time for loops. One signal is that users are already producing something with the product: a shared report, an invited teammate, a published page, a comment left behind. If that output goes to waste today, the raw material for a loop is ready.

It is too early when the product does not stick. If users leave after the first week, a loop will not bring those users back; it pours water into a leaky bucket. Repair retention first.

One distinction matters. The experiment cycle (hypothesis, test, learn) and growth hacking produce tactics; a loop is a system architecture. Experiments are used to improve a single step inside the loop. The loop tells the experiments which question to focus on.

What most teams miss

Teams confuse “going viral” or “adding referral” with building a loop. A referral button is a feature; a loop is when that button forms a chain with the product's core value. If users have to go somewhere unnatural to send an invite, output production sits outside core usage and the loop does not turn.

Dropbox's referral program is the familiar example: inviting was tied directly to what the user was already doing with the product (storage), and the reward was the product's own value. The output was a natural extension of the core behavior.

A loop does not try to persuade the user; it turns the by-product of what the user already does into a new user.

The second missed point is choosing the loop type. There are four general kinds: viral (users bring other users), content (user-generated content draws new users through search or sharing), paid (a share of revenue is converted into advertising) and sales (customer revenue feeds the sales engine). Every product has a natural type; forcing one does not work. Building a content loop in a B2B tool is often weaker than a loop of spread inside a team.

Third is loop speed. However good the conversion, if a cycle takes six months, results take a long time to appear. As cycle time shortens, the growth curve bends faster. Build the one of two candidate loops with the shorter cycle time first.

The difference shows up here: a team looking at the funnel says, “let's raise signup conversion.” A team looking at the loop asks, “after a user gets first value, how long does it take them to bring someone new, and at which step do we lose people?” When the question changes, the to-do list changes too.

The playbook

1. Identify the core action and its natural output

What does the user produce while using the product? A document, a share, an invite, some data, a comment? Is that output visible to others, and does it help them? This question shows your loop type.

2. Draw the loop

Four boxes: input (a new user), action (what the user does), output (what comes out), and re-input (the path from output to a new user). Write an estimated conversion rate for each arrow. If there is an arrow you cannot draw, the loop does not exist yet.

3. Find the weakest link

Measure the loss at each step. Is the rate at which a new user produces output low, the rate at which the output is seen, or the rate at which viewers convert to signups? The weakest link determines the loop's total strength. Improve only that link.

4. Shorten cycle time

After a user gets first value, how long until they produce the output? Move the invite or share moment closer to the first experience. Also shorten the output's time to become a new user: a visible link, one-click signup, pre-filled content.

5. Drive the loop with an experiment

A loop is described by one number: for every hundred new users, how many new users does one turn produce? Above one hundred, the loop sustains itself on its own; below, it carries only part of your growth. Both are valuable, but they are managed differently. Tie step improvements to this number through your experiment cycle (hypothesis, test, learning).

6. Mature one loop before adding another

A second loop is usually added once the first is seen working healthily. Three independent half-built loops scatter resources; a single good loop speeds up both the product's growth and its learning.

What to measure

• Loop multiplier: the number of new users that a hundred new users bring after one turn. Above one hundred indicates self-sustaining growth.
• Cycle time: days from a user's entry to the entry of the new user they brought. A leading indicator.
• Step conversions: the real rate on each arrow in the loop; they show the weakest link.
• Retention of loop-sourced users: how well users who came through the loop stay. If low, the loop is producing low-quality users; lagging but critical.

Pitfalls

Confusing a feature with a loop. Adding a share button is not a loop. The output has to be a consequence of the core value.

A low-quality loop. Invites inflated by rewards bring many fast signups and few lasting users. Look at the retention of loop-sourced users more than the raw count.

Too many loops at once. Not every product carries every type. Find the natural one; a forced loop creates friction and damages the core experience.

How AI changes this

AI can shorten a loop's cycle. Steps that take days, such as producing content, personalizing it and preparing the invitation, can drop to minutes; if production is the loop's bottleneck, the cycle gets shorter and the same time can yield more cycles.

But every cycle can carry a variable cost. In a product with inference cost, each output is a call; as the number of cycles grows, total inference cost grows with it. If the cost per acquired user exceeds that user's value, unit economics get worse. The quality of abundantly produced output can also drop: irrelevant content may get no response from the recipient and can pull down the conversion of the next step.

Rule of thumb. Before scaling, measure cost per cycle and output quality; e.g. if the retention of users coming from the loop is falling, a faster loop is not yet a gain.

Our reading: cycle time alone does not judge the health of a loop. Until speed, cost and quality are read on the same table, it may be too early to grow the loop.

Start this week

1. List everything your users produce with the product and send outside (shares, invites, content, reports), and check whether anyone else actually sees each one.
2. For the most promising output, draw a four-box loop and write the estimated rate for each arrow. Identify the data you need to measure the rates you do not know.
3. Count how many of the new users in the last three months came from an existing user's output. That is your first estimate of today's loop multiplier.