Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Growth Marketing › Growth Fundamentals

How a Growth Loop Compounds

How a growth loop turns customers into more customers: generations, why a loop that cannot sustain itself still multiplies, cycle time, decay and churn.
How it worksintermediate18 min readUpdated

What compounds in a loop, and what does not

The growth loops in practice guide maps a loop as a circle of stages, puts a rate on each arrow and multiplies round the circle to get the loop factor: how many new customers an existing customer produces in each cycle. It reads the result as a rule, that a factor above one compounds on its own while a factor below one decays yet still subsidises other acquisition. The growth loops framework calls the same figure the return per cycle and sets cycle time and decay beside it. Both are right, and both leave the arithmetic between the rule and its result to the reader.

That arithmetic matters because the factor describes a single turn and a business lives through many. A customer bought through paid search invites friends, some of them join, and those friends invite people of their own. Counting only the friends who joined credits the loop with its first generation and ignores the rest. Counting everyone who ever opened a shared link credits it with customers that paid channels or search brought, the flattery the loops guide warns about. Between the two sits what happens in practice: a chain of generations whose size and timing follow from the factor, the cycle time and the way both change as the loop runs.

Not everything in a loop compounds. A paid customer's own first purchase is counted once, and so is the spend that bought them. What compounds is the next customer that purchase leads to, who can lead to another in turn. The loop multiplies customers, and through them their margin; it does not multiply the money that started it.

Step by step: an acquired customer and the generations behind them

Follow one group of customers acquired from outside the loop in the same period, called the seed here, through the loop's stages.

  • The seed arrives. Customers come in through paid media, search, partnerships or any other source that is not the loop itself. They are the only customers in the chain the loop did not produce.
  • Some reach the value moment. Only customers who get value produce anything. The share who do is the loop's first arrow, the activation rate wherever the value moment is defined as activation, and a customer who never activates ends their branch of the chain at once.
  • Activated customers produce output. Invitations, shared work, reviews, content or margin, in whatever amount each produces in a cycle.
  • Output reaches people who are not yet customers, and some of them join. The fraction of that output reaching someone new, and the fraction of those people who join, turn the seed's output into the first generation.
  • The first generation runs the loop. These customers activate, produce and convert at their own rates, which need not match the seed's, and the second generation follows about a cycle later.
  • The chain continues until it fades or fills its circle. Each generation is the one before multiplied by the factor the one before achieved. While the factor stays below the point at which each customer replaces themselves, the generations shrink and the chain fades; at or above it they hold or grow until the people the loop can reach run out.
GenerationWho they areSizeWhen they arrive
SeedCustomers acquired from outside the loopWhatever the channels boughtThe period they were acquired in
FirstCustomers the seed producedThe seed multiplied by the seed's factorAbout a cycle later
SecondCustomers the first generation producedThe first generation multiplied by its own factorAbout a further cycle later
LaterThe customers each generation producesSmaller each time while the factor stays short of self-sustainingA cycle apart, blurred by the spread in cycle times
The whole chainEveryone the seed set off, the seed includedThe seed multiplied by the loop's multipleSpread over every cycle the chain lasts

The last row is the one teams leave out. A loop's value for each customer it starts from is not its factor; it is the whole chain, which this page calls the loop's multiple.

Why a loop that cannot sustain itself still multiplies acquisition

Add up the generations of a loop whose factor is short of self-sustaining and the total settles. Each generation is a fixed fraction of the one before, so each addition is smaller than the last and the sum stops growing. Put in words, and assuming every generation runs the loop at the same factor, the customers the loop produces for each seed customer equal the factor divided by what is left when the factor is taken away from one, and the whole chain including the seed customer equals one divided by that same remainder. Where produced customers run the loop at a different factor, the same addition is done generation by generation with each generation's own factor. Neither needs a figure from outside the business. Both need only the factors the business has measured, which is why the measurement in the loops guide comes first.

Three consequences follow.

A small factor adds little beyond its first generation. When the factor is a small fraction, the later generations are tiny and the customers the loop produces barely outnumber the first generation. A loop map that counts only the first round, as the loops guide's own design tool example does, loses almost nothing by stopping there.

A larger factor puts more of the value into later generations. As the factor approaches the self-sustaining point, the remainder that divides it shrinks, the multiple grows faster than the factor does, and a rising share of the produced customers arrives in the second and later generations. A team that counts only the first round undervalues a strong loop far more than a weak one.

Near the self-sustaining point, small changes move the multiple a long way. Because the multiple depends on the gap between the factor and one, the same gain in the factor is worth much more in a strong loop than in a weak one. Errors grow in the same way: a factor overstated by loose attribution inflates the multiple out of proportion, which is one more reason the loops guide wants the rule for crediting produced customers written down and kept by someone other than the loop's owner.

The multiple is what produces the fall in blended acquisition cost (CAC) that the loops guide's measurement table compares with the cost of paid customers alone. Spend is divided by the seed and by every generation the seed set off, so a loop short of self-sustaining still lowers the blended cost of every customer the business gains, provided the produced customers would not have come anyway.

Cycle time: when the produced customers arrive

The multiple says how many customers a seed sets off; cycle time says when they arrive. Each generation comes about a cycle after the one before, and because cycle times vary from customer to customer, a seed's chain arrives as a spread rather than in steps. The loops guide measures cycle time as the median gap between a customer's acquisition and the acquisition of the first customer they produce; the spread around that median decides how long the tail takes.

Set the cycle against the period the business reports on:

  • A cycle that is short against the reporting period. Most of a seed's chain arrives inside the period, and the period's figures carry nearly the whole multiple.
  • A cycle that is long against the reporting period. Most of the chain arrives after the period closes. The loops guide's referral loop for professional firms, where a satisfied client's peer gets in touch months later, and the slow cycle the framework gives the company content loop are this kind. The period's acquisition cost looks worse than it will turn out, and a change made now shows in the loop's output only several reporting periods later.

Timing produces two further effects.

Steady seeding produces a steady loop. When the business acquires a similar seed every period, each period's loop-produced customers are a mix of first, second and later generations from many earlier seeds. Once enough chains overlap, the loop's output per period settles at the seed per period multiplied by the produced part of the multiple. While the factor stays short of self-sustaining, the cycle time changes how long that level takes to build, not the level itself.

Changes to seeding reach the loop late. Cut the paid seed and the loop's output does not fall at once: chains already started keep delivering for about a cycle in a weak loop and for several cycles in a strong one before they fade. Raise the seed and the extra produced customers arrive after the same delay. Where a loop sits behind paid acquisition, the growth accounting tutorial is the place to watch new customers by source in the cycles after a seeding change.

The loops guide makes the matching point about the current year: a weaker loop that turns fast can deliver more within it than a stronger loop that turns slowly, because compounding depends on both the factor and the cycle. On this page's arithmetic, the slow loop's chains are still arriving when the year closes.

What wears the factor down: activation, churn, saturation and decay

The arithmetic above treats the factor as fixed. In a running loop it moves: four forces push it down, and produced customers can move it either way.

Activation. Customers who never reach the value moment produce nothing, so the share who activate multiplies everything after it. That is why the loops guide's design tool example finds the leak that matters at activation, not at the share prompt the team had been arguing about.

Churn before producing. A customer who leaves before inviting, reviewing or buying again ends their branch of the chain. Early churn therefore lowers the factor as well as the customer's own value, and a loop read without the retention curve beside it cannot say why its factor fell. The retention and payback guide reads that curve.

Saturation. Customers reach new people through the circles they belong to: colleagues, neighbours, a class, a club, the people who search for the same thing. As more of a circle has already joined or already declined, more of each customer's output lands on people the loop has reached before, and the conversion arrow falls. It is diminishing returns applied to the people a loop can reach.

Decay across cohorts. The framework's limits section records that early cohorts invite more and early content ranks more easily, so a factor measured early flatters the long run. The product launch application of the loop framework expects its earliest cohorts, the seed group and the launch week, to behave differently from the cohorts that follow. Produced customers can also run the loop at a different rate from the seed, in either direction, and only reading them as their own cohort shows which.

Each of these shows only when the factor is read by Cohort: the loop's stage rates for the customers acquired in each period, and for each generation where referral data can tell them apart. The cohort analysis tutorial builds that table. A factor read on the whole base averages a young, eager cohort with an old, saturated one and hides the decline. Generations of a small seed are small counts, and their swings from month to month are what the guide to reading small numbers warns against over-reading.

ForceArrow it movesHow it shows in the cohort readingWhere the site treats it
ActivationCustomers reaching valueA smaller share of each cohort producing anythingThe loops guide and its design tool example
Churn before producingOutput per customerCustomers leaving before their first invitation, review or repeat purchaseThe retention and payback guide
SaturationConversion of the people reachedOutput holding up while the customers it produces fallThe framework's loop types table and loop metrics
Decay across cohortsEvery arrow, graduallyLater cohorts' factors sitting below earlier onesThe framework and its launch application
Different produced customersAny arrow, for produced customers, in either directionProduced customers' factor differing from the seed'sThe cohort analysis tutorial, with produced customers read as their own cohort where referral data allows

For apps, the app installs application of the loop framework adds the store page as an arrow of its own, which lowers the factor between an accepted invitation and an installed customer.

Not every loop runs on invitations. Three other kinds follow the same arithmetic with a different unit.

The paid loop. The output is margin. Customers bought with spend return contribution margin over time, and the business reinvests it to buy the next customers. Choose a cycle, such as the period over which spend is planned. The factor is the margin a cohort returns within that cycle divided by the spend that bought it. When the cohort earns back its acquisition cost inside the cycle, in other words when its payback period is shorter than the cycle, each round can fund one at least as large; when payback takes longer, each round funds only part of the next and the rest must come from cash already held. The loops guide's paid row and the LTV and CAC guide set the condition: the paid loop compounds only on margin the business really earns and only if that margin returns before the cash to wait for it runs out. Saturation applies here too, as each further round buys customers at a higher cost.

The marketplace loop. Each side is the other's input: supply that buyers can use draws more buyers, and buyers that sellers can serve draw more sellers. The factor has to be read by area and category, because the two-sided marketplace guide finds growth local and imbalance the main leak: where one side is short, the other side's output converts nothing, and the factor in that area collapses whatever the average says.

Content and product-embedded loops. The output is exposure rather than an invitation, the conversion arrow passes through search results, feeds or a recipient's inbox, and the cycle can be long. The arithmetic does not change; the measurement is harder, which the framework flags for content loops.

What it means for you: valuing a loop, reading acquisition cost and choosing the arrow to improve

The mechanism turns into a few working habits.

Value the loop as a multiple of each seed, by cohort. Report how many customers each seed cohort has produced so far, generation by generation where the data allows, beside the factor. The produced part of the multiple, not the factor, is what the loop adds to each acquired customer, and the cohort reading shows whether it is holding.

Read acquisition cost twice, and date both readings. Keep the paid channels' own cost per customer and the blended cost across seed and produced customers, as the loops guide's table sets out, and state how much of the chain had arrived when each was read. A blended figure read before the tail arrives overstates cost; one that counts an expected tail as if it had arrived understates it. Treat the expected tail the way the LTV and CAC guide treats a lifetime beyond the observed window: shown, labelled as a projection and kept out of decisions until the cohorts have aged.

Credit only what the loop added. Some produced customers would have come anyway, through ordinary word of mouth or search. Where the loop runs through a referral programme, the tutorial on designing one measures referred customers beyond those, and the incrementality testing guide describes holding an area or a group out. A multiple built on customers who would have arrived regardless measures nothing the loop did.

Let measured chains, not first purchases, set what a seed customer is worth. If customers from one channel activate and produce more than customers from another, each sets off a longer chain, and what a seed customer is worth differs by channel even when first-purchase value does not. Read the factor by acquisition channel beside each channel's own acquisition cost, which the LTV and CAC guide keeps as the basis for budget decisions, and base any change to what a channel may pay on chains that have been observed.

Choose the arrow by what a feasible change would gain, not by the lowest rate, as the loops guide's rule for finding the leak that matters already says, and weigh the effort beside it. The factor is a product of rates, so raising any arrow by a given proportion raises the factor by that same proportion. The useful question is which arrow can be raised most for the least work, and the impact-effort matrix is a quick way to sort the candidate changes before they are tested. In a strong loop each gain is worth more, because the multiple responds faster.

Expect the lag, and plan for the circle filling. A change to seeding or to an arrow shows in the loop's output over a cycle or more, and longer in a strong loop; judge it once the chain has had time to arrive, in the same spirit as the experimentation cadence guide revisits shipped wins against cohort data later. Because circles fill, a factor that holds today can fall as the reachable people are used up. The spreadsheet growth model adds a loop's measured output as its own source in the acquisition row rather than assuming it compounds, which keeps that decline in view.

A carpooling app reads what its invitation loop is worth (illustrative)

Illustrative example — not a real result

Plavexund Lifts counts the parents its loop really brings

Plavexund Lifts, an invented carpooling app that parents use to share school and club journeys, charges a family subscription after a free first month. In the first fortnight of the spring term it acquires 800 parents through paid social and adverts in school newsletters for £19,200, a paid cost of £24 a parent. A parent who sets up a shared run invites the other parents on it.

The growth lead maps the loop for the spring cohort: 55% of new parents set up a first shared run within two weeks, each of those sends 2.0 invitations on average, and 35% of invitations are accepted. The seed's factor is therefore 0.55 × 2.0 × 0.35, or 0.385, and the median time from a parent joining to the first parent they invited joining is about three weeks.

Reading the invited parents as their own cohort shows that they run the loop less: many join a run someone else organises, so only 36% set up a run of their own, and their factor is about 0.25. The spring seed's produced parents are therefore 308 in the first generation (800 × 0.385), then 77, 19, 5 and a final handful: about 410 in all, so the whole chain of about 1,210 is a multiple of a little over 1.5 on each paid parent.

ReadingParents countedCost per parent
Paid parents only800£24.00
Month-end report, four weeks in800 plus 190 of the first generation£19.39
End of term, about three cycles after the seed800 plus 404£15.95
Whole chain, with the last 6 still to arrive800 plus 410£15.87

The month-end report had shown only part of the first generation. The team now uses the end-of-term reading as its decision figure and labels the remaining tail as a projection.

The cohort reading also shows saturation. At the 12 schools where the app launched first, acceptance has fallen from 41% to 22% over a year, because most parents there have already joined or declined; at newer schools it is 38%. The team stops treating its average factor as stable and plans its seeding school by school, starting where few parents have heard of the app.

Two changes compete for the next build slot: a preview of the run inside the invitation, aimed at acceptance, and a simpler first-run set-up, aimed at the share of new parents who set one up. Raising set-up from 55% to 66% or acceptance from 35% to 42% would each lift the seed's factor from 0.385 to 0.462, so the team sorts them by effort and starts with set-up, the cheaper of the two to build.

Finally the team checks what it is crediting. A term-long holdout, with in-app invitations switched off at a matched group of schools, shows new parents still arriving there through playground conversation at about a fifth of the rate at which invited parents join at comparable schools, so its budget model counts four-fifths of the parents each chain produces as the loop's addition, which moves the end-of-term decision figure from £15.95 to about £17.09 a parent.

Common mistakes

  • Valuing a loop at its first generation. Produced customers run the loop too; count the chain, and in a strong loop expect the later generations to be a large part of it.
  • Counting the expected tail as if it had arrived. Report what has arrived, date it, and label the rest as a projection until the cohorts have aged.
  • Reading the factor on the whole customer base. Saturation and decay hide in an average; read the factor by cohort and, where the data allows, by generation.
  • Assuming produced customers behave like the seed. Invited customers can activate and invite at quite different rates; measure them as their own cohort.
  • Judging a seeding change in the period it was made. Chains arrive over one or more cycles, longer the stronger the loop, so a cut or a rise in paid acquisition shows in the loop's output late.
  • Crediting customers who would have come anyway. A multiple built on ordinary word of mouth flatters the loop; hold out a group or an area and count only the difference.
  • Running the paid loop on revenue. The factor of a paid loop is margin returned within the cycle; measured on revenue it looks self-funding when it is not.
  • Expecting the factor to hold for ever. Circles fill and early cohorts are keener; plan for the factor to fall and seed new circles before it does.

The growth fundamentals hub places the loop mechanism beside the north star and the cadence of experiments that work on it.

Frequently asked questions

Is a growth loop worth having if each customer brings in less than one new customer per cycle?

Yes, provided the customers it produces are worth what the loop costs to run. The customers a loop brings in run the loop as well, so each acquired customer sets off a chain of generations that shrinks but adds up to more than the first generation alone, and that chain lowers the blended cost of every customer the business gains. What such a loop cannot do is grow the business on its own: without customers acquired from outside, the chains fade.

Why do loop customers keep arriving after paid spend is cut?

Because chains set off by customers acquired earlier are still running. Each generation arrives roughly a cycle after the one before, so customers bought before the cut go on producing new customers for a cycle or more, and for longer the stronger the loop, before the flow fades. The same delay works in the other direction: a rise in paid acquisition shows in the loop's output only once the produced customers have had time to arrive.

Should loop-produced customers lower a paid channel's acquisition cost?

Report two figures rather than replacing one. Keep the channel's own cost per customer so it can still be compared with other channels, and add a blended figure that divides the spend by the channel's customers together with everyone they went on to bring. Date the blended figure, because the chain is still arriving, count only produced customers the business would not have won anyway, and apply a written attribution rule held by someone other than the loop's owner.

Terms used on this page

Frameworks

Related