The growth flywheel: compounding loops instead of linear funnels

A funnel describes one pass, a loop describes what the pass leaves behind

A funnel is the familiar picture of selling. Lots of people at the top, fewer at each step, a few customers at the bottom. It is a good picture of one journey. What it leaves out is everything a customer does after the bottom: tells a friend, writes a review, asks a question that becomes your next article, comes back next month. In a funnel, those things simply fall off the page.

A growth loop is the picture that keeps them. The output of one pass, a new customer, becomes the input of the next pass, because that customer brings in more people. A flywheel is the older, looser word for the same hope: a wheel that is heavy to start and easier to keep turning, because each turn adds to the momentum of the last.

I like the idea. I distrust how it is usually sold, as if drawing a circle on a slide made the circle real. This article is the version I use: what a loop actually is, the one number that tells you whether it is doing anything, why that number is nearly always below one, and why a loop below one is still worth having. The funnel itself is covered in full funnel growth frameworks, so here I start where the funnel stops.

Throughout I use one invented shop. Kettle and Crumb sells loose leaf tea and biscuits online from a unit on an industrial estate outside Galway, at kettleandcrumb.example, and is run by a woman called Clodagh. Nobody real, and every number is made up so I can show every step.

Worked example: the funnel Clodagh already has

In March, Clodagh spends €24,000 on paid search and social. Her ads bring 80,000 visitors, and 1,000 of them place a first order. That is a conversion rate of $\frac{1{,}000}{80{,}000} = 1.25\%$ and a cost per new customer of $\frac{€24{,}000}{1{,}000} = €24$.

The funnel stops there. If Clodagh spends nothing in April, the funnel says she gets nothing in April. Whether that is true depends on what those 1,000 customers do next, and that is the whole subject of this article.

Where the flywheel idea comes from, and what it does not prove

The word comes from Jim Collins. In his book Good to Great he described the companies he studied as pushing a giant, heavy flywheel, turn upon turn, until a breakthrough, and he wrote that when asked for the one big push, the answer was all of them added together, effort applied in a consistent direction.¹ His later monograph, Turning the Flywheel, published in January 2019, names Amazon as a company that consciously harnessed the flywheel effect.²

Amazon's own words are more careful than the legend. In the 2005 letter to shareholders, Jeff Bezos explained a decision to keep lowering prices year after year, returning efficiency improvements and scale economies to customers, because he believed it creates a virtuous cycle that leads, over the long term, to much larger free cash flow. In the same letter he called it a decision that cannot be made in a math based way.³ That is a judgement by a chief executive, stated as a judgement. It is not a measured result, and nobody outside Amazon can rerun it.

The marketing software vendor HubSpot turned the picture into a sales model: attract, engage and delight, with "forces" that speed the wheel up and "friction" that slows it down.⁴ It is a useful way to organise a team's attention. It is also a vendor's model of the world, published by a company that sells the software to run it, and the page offers no study showing that firms which adopt it grow faster.

The most useful version for practical work came from Brian Balfour, Casey Winters, Kevin Kwok and Andrew Chen in 2018, who described growth loops as closed systems where the inputs, through some process, generate more of an output, and who put the focus on how you reinvest the output of one cycle into the next.⁵ That definition has a virtue the others lack. It can be counted.

An example: from Collins, Clodagh gets a mood, from Bezos a conviction, and from the loop definition a question she can answer from her own data. Of the 1,000 March customers, how many new customers did they bring in, and how long did it take? The rest of this article answers that.

The one number that matters: the gain of the loop

Every loop has a gain. The gain is how many new customers one new customer brings in during one turn of the loop. For a referral loop it breaks down into three parts:

$$g = s \times i \times c$$

Here $g$ is the gain of the loop, the new customers produced per new customer per cycle. $s$ is the share of new customers who take the loop action at all, for example sharing a referral link. $i$ is the number of people each of them reaches, for example friends who open the link. $c$ is the conversion rate of those people into customers. Each of the three is a plain rate, events divided by opportunities, as covered in Foundations: ratios and rates.

This is the same arithmetic David Skok published for viral products, where the viral coefficient $K$ is the number of invitations each customer sends multiplied by the conversion rate of those invitations.⁶ I split out $s$ because in a shop most customers never share anything, and that silent majority is where most loops actually fail.

The old edition of this article offered a "flywheel strength" made of output over input multiplied by a retention rate. I have retired it. Output over input already contains whatever retention did, so multiplying by retention again counts it twice, and the result had no unit you could check against your books. The gain $g$ has a plain meaning and comes straight out of an order export.

Worked example: the gain of Clodagh's referral loop

After every first order, Kettle and Crumb's thank you page and dispatch email offer a link: give a friend €5 off, get €5 off yourself. In March, 250 of the 1,000 new customers shared it, so $s = \frac{250}{1{,}000} = 0.25$. Each sharer's link was opened by four friends on average, so $i = 4$. Of those 1,000 friends who opened a link, 200 placed an order, so $c = \frac{200}{1{,}000} = 0.20$.

The gain is $g = 0.25 \times 4 \times 0.20 = 0.20$. Every 100 new customers bring in 20 more. Grand for a shop, and nowhere near the self propelling machine of the slide decks. That matters less than you would think.

Below one, a loop multiplies your spend. Above one, it runs on its own

Follow the 1,000 March customers through the loop. They bring in 200. Those 200 bring in 40. Those 40 bring in eight, and so on, each wave a fifth of the one before. The waves shrink, and they add up to a fixed total:

$$N_{\text{total}} = N_0\,(1 + g + g^2 + g^3 + \dots) = \frac{N_0}{1 - g} \quad \text{for } g < 1$$

Here $N_{\text{total}}$ is the total number of customers that one starting group eventually produces, itself included. $N_0$ is the starting group, the customers you paid for. $g$ is the gain from above, and $g^2$, $g^3$ are the second and third waves, the friends of friends. The sum of a series that shrinks by the same factor each step is the same multiplying logic that underlies compound growth, explained in Foundations: compounding.

The factor $\frac{1}{1-g}$ is the loop multiplier. It is the honest answer to "what is my loop worth". Skok's own example lands on it: ten starting customers with a viral coefficient of 0.5 grow to 20 people and then stop, exactly $\frac{10}{1 - 0.5}$.⁶ Skok's conclusion follows from the same sum: the coefficient must be greater than one for true viral growth.⁶ Above one, the series never settles, and every wave is larger than the last.

In the shops I have worked on I have never measured a gain above one, and I would be suspicious of anyone selling a loop for a physical product that claims one. Below one is normal. Below one is also fine, as the next example shows.

Worked example: what the loop does to Clodagh's cost per customer

Clodagh keeps spending €24,000 a month and keeps buying 1,000 customers a month at €24 each. Referrals from one month's customers arrive the following month, so the cycle is one month long. New customers per month then run:

• Month one: 1,000 paid.
• Month two: 1,000 paid plus 200 referred from month one, 1,200.
• Month three: 1,000 paid plus $0.20 \times 1{,}200 = 240$ referred, 1,240.
• Month four: 1,000 plus $0.20 \times 1{,}240 = 248$, 1,248.
• Month five: 1,000 plus $0.20 \times 1{,}248 = 249.6$, about 1,250.

The total settles at $\frac{1{,}000}{1 - 0.20} = 1{,}250$ a month. The cost per new customer settles at $\frac{€24{,}000}{1{,}250} = €19.20$, which is the paid cost multiplied by $(1 - g)$, so $€24 \times 0.80$.

Now the honest correction. The loop is not free. Each referred customer costs €10 in rewards, €5 to the friend and €5 to the sharer. That is 250 times €10, €2,500 a month, so the true cost per new customer is $\frac{€24{,}000 + €2{,}500}{1{,}250} = €21.20$. Still lower than €24, and lower every month she runs it. But it fell once, to a floor, and stayed there. The old claim that flywheels reduce customer acquisition cost (CAC) over time is half true: a loop lowers the cost of acquiring a customer to a new level, and then it holds. It does not keep falling unless the gain itself keeps rising.

Across the bottom are the months, along the side the number of new customers that month, paid and referred together. Both lines climb and then go flat. The flat level is $\frac{1{,}000}{1-g}$: 1,250 with a gain of 0.2, 2,000 with a gain of 0.5. The takeaway: a loop below one raises the ceiling of your spend, it does not remove the ceiling. Raising the gain from 0.2 to 0.5 is worth far more than any month of patience.

Cycle time decides how fast you get there

The gain decides how far a loop takes you. The cycle time, the time it takes a new customer to bring in the next one, decides how quickly. Skok makes the point with a viral product: after 20 days with a cycle time of two days he gets 20,470 users, and halving the cycle time to one day gives over 20 million.⁶ That is the arithmetic of a gain above one, where every cycle multiplies the base, so doubling the number of cycles squares the multiplier.

Below one, cycle time matters less dramatically and more practically. It does not change the ceiling of $\frac{N_0}{1-g}$. It changes how long the referred customers take to arrive, and money that arrives sooner is worth more than money that arrives later.

Worked example: Clodagh shortens the cycle

Clodagh notices that most March sharers shared only when they reopened the dispatch email, about four weeks after ordering. She adds the referral offer to the parcel itself, on a card in the box, and the typical gap drops to two weeks.

The gain stays 0.20, so the monthly total still settles at 1,250. What changes is the first month of any new push. If she launches in a new region with 1,000 paid customers, the 200 referred customers now arrive within the same month rather than the next. Her first month's cost per customer, rewards included, is $\frac{€24{,}000 + 200 \times €10}{1{,}200} = €21.67$ straight away, instead of €24 and a month's wait. For a shop that pays suppliers before it is paid, that is a cash flow improvement first and a growth improvement second.

When a loop compounds, and what the old maths left out

The previous edition of this article said that a loop adding 10% more users per cycle, run 12 times a year, gives $(1.10)^{12} = 3.14$, "3x growth with no additional spend". The arithmetic is right. The claim is not. Here is the formula, then the assumptions hiding in it.

$$B_n = B_0\,(1 + g)^n$$

Here $B_n$ is the customer base after $n$ cycles, $B_0$ is the base at the start, $g$ is the growth per cycle as a fraction, and $n$ is the number of cycles. It is compound growth, as in Foundations: compounding, and its trap is the same one: you cannot add percentages across periods, 10% a month for a year is about 214% growth, not 120%.

For this to hold, every customer in the base, old and new, has to keep producing new customers at the same rate every cycle, forever, and nobody may leave. That is a different loop from the one in the last section, where only the newest customers referred. Real loops sit between the two. A customer refers most in the first weeks, a little later, and not at all after they lapse.

And nothing in it is free. Referral rewards, review requests, the time to write the next article: every turn of a loop costs something. "No additional spend" is the part of the old claim I would never put my name to.

Worked example: Clodagh's base with churn in it

Suppose, generously, that Clodagh's whole base of 5,000 customers does bring in 10% more each month. She also loses 3% of her customers each month, people who simply stop ordering. The net growth per cycle is $0.10 - 0.03 = 0.07$.

After a year the base is $5{,}000 \times 1.07^{12} = 5{,}000 \times 2.252 = 11{,}261$, against $5{,}000 \times 1.10^{12} = 5{,}000 \times 3.138 = 15{,}692$ without churn. Three points of monthly churn cost more than 4,400 customers by December. Which is why the next section is about the thing loops quietly depend on.

Retention is the loop's memory

A customer who stays has more chances to turn the loop. They reorder, which is margin. They pass the link on again when a colleague mentions tea. They write the second review, the one that says the subscription still arrives on time a year later. A customer who leaves after one order takes all of that with them.

The value of a customer who stays is their lifetime value. With a constant margin and retention, it has a short form:

$$CLV = M \times \frac{r}{1 + d - r}$$

Here $CLV$ is the customer lifetime value, $M$ is the margin a customer brings per year (margin, not revenue), $r$ is the retention rate, the share of customers still buying a year later, and $d$ is the discount rate, which turns future money into today's money. Each symbol, and why margin and not revenue goes in, is explained in Foundations: lifetime value.

The same retention rate also tells you how long a customer stays in the loop. The expected number of years a customer stays is:

$$T = \frac{1}{1 - r}$$

Here $T$ is the expected customer lifetime in years and $r$ is the yearly retention rate from above. It is the same shrinking series as the loop multiplier, with retention in place of gain, so Foundations: compounding covers it too.

Worked example: what retention is worth to Kettle and Crumb

Clodagh's customers bring €36 of margin a year. With retention at 55% and a discount rate of 10%:

$$CLV = 36 \times \frac{0.55}{1 + 0.10 - 0.55} = 36 \times \frac{0.55}{0.55} = €36$$

Here $36$ is the yearly margin $M$ in euro, $0.55$ is the retention rate $r$ and $0.10$ is the discount rate $d$, exactly as in Foundations: lifetime value. At 70% retention it is $36 \times \frac{0.70}{0.40} = €63$, and at 80% it is $36 \times \frac{0.80}{0.30} = €96$. The expected lifetime moves from $\frac{1}{0.45} = 2.2$ years at 55% to $\frac{1}{0.20} = 5$ years at 80%.

So a customer who costs €21.20 to acquire is worth €36 at 55% retention, and €96 at 80%. And at 80% each customer spends five years in the loop rather than two, with five chances to share rather than two.

Across the bottom is the yearly retention rate, up the side the lifetime value of one customer in euro. The line is not straight, it bends upward: each ten points of retention is worth more than the ten before, from €30 at 50% to €162 at 90%. The takeaway: the last stretch of retention is the most valuable stretch, and it is also the stretch that feeds the loop longest. Survival curves and churn are treated in depth in survival analysis, churn and customer lifetime value.

Three loops a shop can actually run, and the rules on each

Most shops have three loops within reach: referrals, reviews and content. Each one has a weak link, and in Ireland each one has a rule worth knowing before you build it.

Read from the top: the first decision is $s$, the share who share, and the second is $c$, the friends who buy. The number of friends each sharer reaches, $i$, sits on the arrow between them. Each box is a place where the loop can leak, and each decision is a number Clodagh can measure.

The referral loop and the email rules

The referral loop is the one with the cleanest gain. Its trap is the tempting shortcut: asking customers to type in their friends' email addresses so the shop can send the invitation. In Ireland, marketing email is governed by regulation 13 of the ePrivacy Regulations 2011 (S.I. No. 336 of 2011), which requires the recipient's consent unless they are already your customer.⁷ The Data Protection Commission (DPC) has answered this exact question. Sending an unsolicited marketing message to someone with no existing customer relationship is an offence, the onus is on the sender to show the recipient consented, and, in the DPC's words, it is difficult to see how the operation of a "refer a friend" facility could meet this requirement.⁸

An example: Clodagh's card gives the customer a personal link, kettleandcrumb.example/r/orla5, to pass on however they like, in a message of their own. The shop never holds the friend's address until the friend chooses to order. That keeps the loop on the right side of the DPC, and in my experience a personal message converts better than a shop email anyway.

The review loop and the fake reviews rules

The review loop runs: an order arrives, the customer is asked for a review, the review appears on the product page, the next visitor buys with more confidence, and that visitor is asked in turn. Google can show review stars for products in search results, but it states that it does not guarantee that features using structured data will show up, and that pages using LocalBusiness or Organization markup are not eligible for review stars about themselves.⁹

The rules are stricter than most shops realise, and in Ireland they are set at EU level. The Omnibus Directive added two practices to the list that is banned in all circumstances: claiming that reviews come from people who actually bought or used the product without taking reasonable and proportionate steps to check, and submitting or commissioning false reviews, or misrepresenting reviews, to promote a product. It also made it material information whether and how a business checks that its reviews come from real buyers.¹⁰ Ireland wrote both into the Consumer Protection Act 2007 through the Consumer Rights Act 2022, in sections 46(3B) and 55(1)(za) and (zb),¹¹ and the Competition and Consumer Protection Commission (CCPC) enforces them.¹² Nothing in that list stops a shop rewarding a review with a discount or points. The European Commission's guidance on the Directive is plain about the two ways it goes wrong. It names traders who incentivise consumers to test their products in exchange for posting their reviews "without disclosing the fact of sponsorship", and it says that "Misleading practices regarding consumer reviews and endorsements may breach Article 7(2) UCPD", the rule against hiding a practice's commercial intent. It also counts "only soliciting and making available positive reviews" as misrepresenting reviews.¹³ So the safe loop is simple: ask everyone, and label every reward.

An example: Clodagh emails every customer ten days after delivery, happy or not, with the same request. 1,250 orders a month and a 10% response give 125 reviews a month. If she adds 50 loyalty points for any review, positive or negative, each of those reviews carries a visible "received loyalty points" label. That is a loop she can defend in front of the CCPC.

The content loop, and what it really feeds

The content loop runs: a customer asks a question, the answer becomes an article, the article brings searchers, some searchers buy, and their questions feed the next article. The old edition of this article claimed that shares and engagement boost rankings. I cannot support that. Google's own description of how search works says relevance is determined by hundreds of factors and gives the searcher's location, language and device as examples. It does not name social shares.¹⁴

So I treat content as a learning loop first. The new customers it brings are real but slow, and the gain is hard to measure. What it reliably compounds is understanding: every question answered is one less objection at the checkout.

An example: Clodagh's customer service inbox logs 30 distinct questions in a quarter. She turns the ten most frequent into short guides, such as how to brew a first pot of loose leaf tea. Over the next quarter the guides bring 6,000 visits from search, 1.5% of which, 90 visitors, place an order. Their emails produce 12 new questions, and those become the next guides. The 90 orders are welcome. The 12 questions are the part that keeps turning.

How to measure a loop without fooling yourself

A loop that is not measured is a circle on a slide. Measuring one needs three things: a way to know who came from whom, a cohort to follow, and enough time.

Who came from whom is the referral code, the review request ID or the article a buyer landed on, stored with the order. The cohort is a group of customers who joined in the same period, followed over the months after. Group by cohort, count the new customers each cohort produced, and divide by the cohort's size. That is the measured gain. The median days between a customer's first order and the first order they produced is the measured cycle time. The wider practice of cohort analysis is in what data science actually delivers for growth.

Two traps. First, a referred customer who would have bought anyway is not produced by the loop, so a measured gain is an upper bound until you compare it with the word of mouth you had before the loop existed. Second, a gain measured on 30 orders is noise, not a trend. Wait for enough customers before you believe the number.

Worked example: Clodagh's March cohort

Clodagh tags the 1,000 March customers and follows their referral codes:

• April: 140 new customers used a March code.
• May: 40 more.
• June: 10 more.

The measured gain over three months is $\frac{140 + 40 + 10}{1{,}000} = 0.19$. Most of it arrived in the first month, so the cycle time is about a month, and the loop is concentrated in the early weeks, as expected.

Then the comparison. Before the card existed, Clodagh's checkout asked new customers how they heard about the shop, and in the autumn about 60 new customers a month answered "a friend", against roughly 1,000 paid ones. Word of mouth already had a gain of about $\frac{60}{1{,}000} = 0.06$. The loop's own contribution is therefore roughly $0.19 - 0.06 = 0.13$, not 0.19. It is a rough figure, because autumn is not spring, but it is the number that belongs in her budget, and a test that offers the card to a random half of new customers would tighten it.

When a funnel is the honest answer

Some businesses have no loop worth building. A product bought once in a decade, a service nobody talks about at dinner, a market where customers will not recommend a supplier to a competitor: the gain will sit near zero whatever you draw on the whiteboard. For those, a well run funnel with good retention is the right machine, and calling it a flywheel changes nothing.

For everyone else, I do the same three things. Measure the gain and the cycle time of each loop you already have, because you almost certainly have one and nobody has counted it. Raise the gain at its weakest link, which is nearly always the share of customers who take the first action. And keep the customers you have, because retention is what keeps every loop supplied. A loop with a gain of 0.2 will not make anyone famous. It will quietly make every euro of marketing go a fifth further, for as long as the shop runs it, and that is a flywheel I can stand behind.

Footnotes

1. Jim Collins, The flywheel effect, jimcollins.com, accessed 2026, https://www.jimcollins.com/concepts/the-flywheel.html

2. Jim Collins, Turning the Flywheel: a monograph to accompany Good to Great, HarperCollins, January 2019, https://www.jimcollins.com/books/turning-the-flywheel.html

3. Jeffrey P. Bezos, 2005 letter to shareholders, Amazon.com, Inc., exhibit 99.1 filed with the US Securities and Exchange Commission, 2006, https://www.sec.gov/Archives/edgar/data/1018724/000119312506084815/dex991.htm

4. HubSpot, The flywheel model, HubSpot, accessed 2026, https://www.hubspot.com/flywheel

5. Brian Balfour, Casey Winters, Kevin Kwok and Andrew Chen, Growth loops are the new funnels, Reforge, 31 July 2018, https://www.reforge.com/blog/growth-loops

6. David Skok, Lessons learned: viral marketing, For Entrepreneurs, accessed 2026, https://www.forentrepreneurs.com/lessons-learnt-viral-marketing/

7. Government of Ireland, European Communities (Electronic Communications Networks and Services) (Privacy and Electronic Communications) Regulations 2011, S.I. No. 336 of 2011, regulation 13, Irish Statute Book, 2011, https://www.irishstatutebook.ie/eli/2011/si/336/made/en/print

8. Data Protection Commission, Can I ask my customers to provide the electronic contact details of friends as part of a "refer a friend" scheme so that I can market them about my products and services?, FAQs on direct marketing, DPC, accessed 2026, https://www.dataprotection.ie/en/faqs/direct-marketing/can-i-ask-my-customers-provide-electronic-contact-details-friends-part-refer-friend-scheme-so-i-can

9. Google, Review snippet (Review, AggregateRating) structured data, Google Search Central, accessed 2026, https://developers.google.com/search/docs/appearance/structured-data/review-snippet

10. European Parliament and Council, Directive (EU) 2019/2161 (the Omnibus Directive), Article 3, inserting Article 7(6) and Annex I points 23b and 23c into Directive 2005/29/EC, Official Journal of the European Union, 2019, https://eur-lex.europa.eu/eli/dir/2019/2161/oj/eng

11. Law Reform Commission, Consumer Protection Act 2007, revised, sections 46(3B) and 55(1)(za) and (zb), as inserted by the Consumer Rights Act 2022, Revised Acts, https://revisedacts.lawreform.ie/eli/2007/act/19/revised/en/html

12. Competition and Consumer Protection Commission, CCPC welcomes new rights for consumers, CCPC, 28 November 2022, https://www.ccpc.ie/news-and-media/news/article/2022/11/28/ccpc-welcomes-new-rights-for-consumers

13. European Commission, Guidance on the interpretation and application of Directive 2005/29/EC concerning unfair business to consumer commercial practices in the internal market, section 4.2.4 User reviews, Official Journal of the European Union C 526, 29 December 2021, https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52021XC1229%2805%29

14. Google, In depth guide to how Google Search works, Google Search Central, accessed 2026, https://developers.google.com/search/docs/fundamentals/how-search-works