All 29 modules are open from the start — nothing here is
locked, and nothing costs anything. Sign in so your progress, titles and
credentials stay with you, on every device you use.
Free forever, with your Google account. No password, no payment.
A funnel is linear, pour users in the top, some fall out the bottom, and you must keep refilling forever. A growth loop feeds the output of one cycle back in as the input to the next, so the product helps grow itself. This guide covers how loops work, their common types, and why they compound where funnels only leak.
Ready?
1
Loops vs. Funnels
Two shapes describe almost all growth, and telling them apart tells you what your numbers are worth.
A funnel is a bucket you pour into. People enter at the top, some reach the bottom, and the process ends there. Growth scales with how much you keep pouring in, which means it scales with money and effort — and stops when they stop.
A loop is a fire. It consumes what it produces: users take an action, that action produces something, and that something brings the next user in at the top. Each turn makes the following turn slightly easier.
The test that separates them is one question: what happens when you stop spending? A funnel empties. A loop keeps turning, more slowly, and can be reignited.
The reason this matters is that the two are indistinguishable while they are working. A company hitting its numbers through paid acquisition and one hitting the same numbers through referrals produce the same line on the same slide, and completely different businesses underneath it.
Both are legitimate, and most companies run several of each. But only one compounds and only one survives a budget cut — so knowing which line you are looking at tells you whether this quarter’s growth will still be there next year.
Linear
Funnel
One-way. Needs constant external refilling at the top. Asks: "how do we get more in?"
Circular
Growth loop
Output reinvested as new input. Compounds. Asks: "how does each user generate the next?"
Both branches describe a chart going up, which is why the question has to be hypothetical rather than observational. Nothing you can measure this quarter separates them — the difference only shows up on the month you stop paying, which is the worst possible moment to find out.
Everyday example
A wood stove and a bonfire both give off heat. The stove keeps going because someone keeps putting logs in; the bonfire, once it is large enough, dries the wood at its own edges and feeds itself for a while. From across a field on a cold night they look identical — and only one of them is still burning in the morning.
The channel that stopped
A company built its growth on paid search and hit its numbers for two years. When the budget was cut in a downturn, sign-ups fell by roughly the amount that had been bought — there was nothing underneath. Nothing about the funnel had been wrong; it simply had no mechanism for the output to become the next input.
Quick check
What's the key structural difference between a funnel and a growth loop?
2
Anatomy of a Loop
Every growth loop has the same four-part skeleton, and being able to name each part is what separates a real loop from a diagram drawn as a circle.
A new user arrives. They take an action — post something, invite someone, create a page. That action produces an output that exists in the part: a public page, an invitation, a review. And that output brings the next user, who enters at step one.
Step four is the whole test. Plenty of processes are drawn as circles and are really funnels with a decorative arrow: if the output does not actually cause new input, the circle is a picture rather than a mechanism.
Naming the steps also tells you where to look when the loop is slow, because the loop only runs as fast as its slowest step. One team mapped a content loop — user arrives, creates a public page, page gets indexed, search visitor lands, some of them create pages — and found indexing took eleven days. That single step set the speed of everything else, no matter what they did to the other three.
So the practical exercise is to write your loop out in four sentences and put a number on each arrow: how many people, how often, how long. A loop you cannot write in four sentences is usually not a loop, and the arrow you cannot put a number on is usually the one that is broken.
The test of a loop is step four. If the output does not feed the input, it is a funnel with a nice story — funnels need refilling from outside, loops reinvest what they produce.
Everyday example
A compost heap: kitchen scraps go in, compost comes out, the compost grows vegetables, and the vegetable scraps go back in. The test of a loop is whether the output re-enters at the top, not whether the diagram is drawn as a circle.
A content loop, step by step
Input: a new user arrives. Action: they create a public
playlist. Output: that playlist gets shared and indexed by search
engines. Reinvest: new listeners discover it, arrive, and some create
playlists of their own, which is the input again.
The output (new users) circles back to produce more of the input. That closure is what
separates a loop from a one-way funnel.
A content loop, one turn at a time
A user arrives, creates a public page, the page is indexed, a search visitor lands on it, and some of them create pages of their own. The team could measure each step, and found step three — indexing — took eleven days, which set the speed of the entire loop no matter what they did to the others.
Quick check
In "user creates content → content gets found in search → new users arrive → some create content," what makes this a loop rather than a funnel?
3
Three Common Loop Types
A viral loop burns invitations. A user invites others because the product is better with them in it, or because they are rewarded for it. It is fast when it works and it saturates: eventually the average user has invited everyone they plausibly know.
A content loop burns pages. Users create something public, search engines index it, visitors arrive and some become creators. It is slow to start and compounds unusually well, because the pages you made two years ago are still working.
A paid loop burns margin. Revenue from existing customers funds the acquisition of new ones. It is the most controllable and the most fragile, because it depends on the arithmetic staying favourable — if the cost to acquire rises above what a customer is worth, the loop runs backwards and does so quietly.
Most durable companies run more than one. The point of naming yours is that the three fail differently: a viral loop saturates, a content loop decays if the content stops being good, and a paid loop inverts. Knowing which failure is coming is what lets you see it early, rather than discovering it in a quarter where the number simply stops moving.
Viral loop
Users bring in other users directly, invites, referrals, or collaboration that exposes non-users to the product.
Content loop
Users (or the product) generate content that attracts new users via search or sharing, some of whom generate more content.
Paid loop
Revenue from users funds acquisition of more users, whose revenue funds more acquisition, only sustainable while a customer earns back more than they cost.
Read the bottom line of each column: saturation, decay, inversion. Knowing which one is coming for you is the entire reason to name your loop — all three look like healthy growth right up until they don’t, and only the third one can run backwards.
Everyday example
Three fires, three fuels. A viral loop burns invitations, a content loop burns pages that rank, a paid loop burns margin recycled into ads. Knowing which one you are running tells you what to protect when things get tight.
The content loop that compounded
A review site's loop ran: visitors arrive from search, some leave reviews, the reviews become pages, the pages rank, and more visitors arrive. Each turn made the next turn slightly easier, and none of it was paid for. The same company's paid channel needed the same money every month to produce the same result.
Quick check
A company reinvests subscription revenue into paid ads that acquire more subscribers, whose subscription revenue funds still more ads. Which loop type is this?
4
Why Loops Compound (or Stall)
Loops compound for the same reason compound interest does: each turn adds to the base that the next turn works on. That is what makes them worth building, and it is also what makes their failure so hard to notice.
The health of a loop is really one number — how many new users each turn produces per user who entered it. Above one, it grows on its own. Below one, it decays no matter how good each individual step looks. And that number changes over time, usually downward, for reasons that have nothing to do with the quality of the work.
The most common of those reasons is saturation. An invite loop slows sharply once the average user has invited everyone they know who would plausibly want the product. The loop has not broken — it has run out of fuel. One team spent two quarters polishing the invite screen before anyone noticed that the screen was not the constraint; the addressable network was.
The second is a step quietly turning into a withdrawal. If the output of your loop starts being spent elsewhere — the content gets paywalled, the referral reward is cut to save money — the loop stops feeding itself and becomes a funnel with history.
The useful diagnostic is to measure each step of the loop separately over time, not the total. A loop that is slowing will almost always show it in one step first, months before the headline number admits anything is wrong.
Cycle time
How long one full turn of the loop takes. Shorter cycles compound faster.
Loop factor
New inputs produced per cycle. Above 1 and the loop grows; below 1 and it shrinks.
Friction
Drop-off at any step chokes the whole loop; the weakest step caps it.
Loops run in reverse too: if the loop factor drops below 1, decline feeds decline.
Nothing is broken on this chart. The loop is turning exactly as designed and the average user has simply run out of people to invite. One team spent two quarters on the invite screen before noticing the screen was never the constraint — which the headline number could not have told them, and one step of the loop, measured on its own, would have.
Everyday example
A savings account only compounds while the interest stays in it. Withdraw the interest each month and you have an income, not growth — the balance never moves. Nothing has broken, and nothing is accumulating either.
Where the loop stalled
An invite loop slowed sharply once the average user had invited everyone they knew who would plausibly want the product. The loop had not broken — it had saturated its fuel. The fix was a new fuel source, not a better invite screen, and the team spent two quarters polishing the screen before they saw that.
Quick check
A viral loop where, over time, each user brings in an average of 0.8 new users. What happens to the loop?
Drill what you learned
Scenario 1
easy
A team's entire growth model is: buy ads → users land → some convert → buy more ads with the marketing budget, sized by a fixed quarterly plan.
Is this a loop or a funnel, and why?
Scenario 2
easy
A design tool grows because users share editable project links with collaborators, who sign up to edit and then share their own projects.
Which loop type is this?
Scenario 3
easy
A blog-style product grows as users publish public posts that rank in search, drawing readers who sign up and publish their own posts.
Which loop type, and what's the reinvestment?
Scenario 4
easy
A viral loop currently has each user bringing in an average of 1.3 new users per cycle.
What does a loop factor of 1.3 imply?
Scenario 5
easy
Two viral loops have the same loop factor of 1.2, but one completes a cycle in 2 days and the other in 30 days.
Which grows faster, and why?
Scenario 6
medium
A PM wants to improve a content loop but every step is roughly equal except publishing, where 90% of users drop off before ever creating a post.
Where's the leverage?
Scenario 7
medium
A subscription business reinvests revenue into ads, but its customer acquisition cost has risen above what an average customer pays over their lifetime.
What's happening to this paid loop?
Scenario 8
medium
A founder describes their growth as a "funnel" and asks only "how do we get more visitors to the top?"
How would loop thinking reframe the question?
Scenario 9
medium
A team adds a referral feature, but users have such a poor first experience that few stay long enough to refer, and those referred also churn fast.
Why isn't the viral loop working?
Scenario 10
medium
A product's growth has quietly reversed: usage and new signups both shrink a little more each month.
How might a loop explain this?
Scenario 11
hard
A PM claims their SEO strategy is automatically a "growth loop" because it brings in users from search.
Is bringing users from search enough to make it a loop?
Scenario 12
hard
A team wants to prioritize between two improvements: one raises the loop factor from 1.1 to 1.3, the other has no effect on the loop.
Which is more strategically valuable for compounding growth?
Scenario 13
hard
A viral loop's invite step works, but invited users land on a confusing signup page and most never activate.
How does this cap the loop?
Scenario 14
hard
Leadership asks why a competitor with a strong content loop keeps pulling ahead despite spending less on ads.
What's the likely explanation?
Scenario 15
hard
A marketplace grows because more buyers attract more sellers, and more sellers (with more selection) attract more buyers.
How should you describe this structure?
Drilled it. Now apply it to a real situation.
Put it to work
From lesson 1
Dropbox stopped buying users and started earning them
Dropbox
Early Dropbox tried paid search and found the economics impossible. Drew
Houston has said publicly that they were paying in the region of
$200-300 to acquire a customer for a $99 product.
The replacement was the double-sided referral — free storage for the
person inviting and the person joining — which is widely reported to
have driven signups up dramatically, with referrals accounting for a
large share of new users at its height.
Why did the referral programme work where paid acquisition could not?
What actually happened
The distinction generalises. Paid acquisition is a funnel — money in,
users out, and it stops the day the money does. Referral was a loop,
where users produced users, so the same spend compounded rather than
repeating.
A funnel converts what you put in. A loop turns its own output back into
its input, which is the only structure that grows without being fed.
From lesson 2
Naming the four parts of a loop
A freelance marketplace
The team believes they have a loop: freelancers who complete a job get a
public profile with reviews, those profiles rank in search, and searchers
sign up. Growth has been flat for two quarters.
Put the loop's four parts in the order they run, from the input to the thing that feeds back.
Drag the rows, or use the arrows, then check.
A new freelancer joinsThe input. Every loop starts with a unit entering it — and the loop is only real if the last step produces more of exactly this.
They complete jobs and collect reviewsThe action. This is the work the product exists for, and the step most likely to be the real bottleneck.
Their profile becomes an indexed page that ranks for their specialityThe output. Something durable that exists in the part afterwards and keeps working without further effort.
Searchers land on that page and some joinThe feedback. This is the step that closes the loop — and if it produces buyers rather than freelancers, the loop does not actually close.
What actually happened
Written out, the flaw was obvious: the loop takes in freelancers and
produces buyers. Buyers are valuable, but they are not the input, so
nothing compounds — it is a funnel with an unusually good top.
The team added the missing return path: a buyer who hires successfully is
prompted to invite the specialists they already work with offline. Only
then did each turn produce more of what it consumed.
Write the loop as input, action, output, feedback. If the feedback does
not produce more of the input, it is not a loop.
From lesson 3
Three loops, three different jobs
A recipe app
Three proposals are on the table for next quarter. Each is a genuine loop,
and they behave completely differently.
The three proposals
Proposal
How a turn works
Turn time
Shared shopping lists
A cook shares a list; the recipient signs up to open it
Days
Recipe pages indexed by search
Published recipes rank; searchers arrive and some sign up
They ran the shared list first — fastest to prove, and cheapest to
abandon if the numbers came back flat. It worked, which funded the
patience the content loop needed.
Loop type determines turn time, and turn time determines how long you
must wait before the evidence means anything.
From lesson 4
Why the loop stopped compounding
A photo-sharing app
The invite loop worked for eighteen months and has now flattened. Nothing
was changed. Here is what each stage of a turn looks like now against a
year ago.
Per 100 new users
Stage
A year ago
Now
Send at least one invite
38
31
Invites sent per sender
4.1
3.8
Invites accepted
22%
9%
New users produced
34
10
Work out why it stalled.
Step 1 of 3
Which module explains most of the collapse from 34 to 10?
So the question becomes: why would the same invitation, sent by the same kind of person, be accepted less than half as often?
Step 2 of 3
Nothing about the invite changed. What is the most likely explanation?
Saturation is not a bug. It is what every loop does eventually, and the only question is whether you notice before or after growth flattens.
Step 3 of 3
What do you do about it?
Loops decay. Planning the next one before the current one flattens is the difference between a growth curve with steps in it and one that plateaus.
What actually happened
They built a second loop on a different pool entirely: shared albums for
events, which reach people who would never have been in anyone's contact
list. Acceptance on that path opened at 27% because it was a fresh
network, not a harder push on an exhausted one.
Every loop saturates. Watch acceptance rather than volume, because
acceptance falls first and volume only follows once it is too late to
respond calmly.