Loops vs funnels
The funnel is the oldest model in marketing and still the most useful for one job: describing how a population narrows from awareness to purchase so that the leak at each stage can be measured. Its weakness is what happens after the bottom. The funnel empties, and filling it again costs what it cost last time, or more as the easiest people are used up. Growth under a funnel model is linear in spend.
A loop closes that gap by asking what a customer produces once acquired, and whether that output can be routed back to the top. Three kinds of output do the job. A customer can produce exposure: content, reviews, shared results, or a visible product that others see. A customer can produce invitations: a direct ask to someone they know, usually prompted by the product at a moment of satisfaction. And a customer can produce margin: money that, reinvested in acquisition, buys the next customer. Each of these becomes an input, and a loop whose output per cycle exceeds the input it needs compounds; one whose output falls short decays, but still lowers the cost of every customer bought through other means.
The difference is not that loops are modern and funnels are old. A loop is a funnel drawn with its ends joined, and every stage inside the loop is still a conversion step with a rate (CVRShare of visits or clicks that convert.) that can be measured and improved. What changes is the question the team asks. Under a funnel the question is "how do we fill the top more cheaply"; under a loop it is "how do we make each turn produce more of the next turn". The growth fundamentals hub sets this in the context of the growth model as a whole, and the growth versus traditional comparison there explains why loop thinking changes the shape of the team. At the level of the whole business rather than a single loop, the comparison of the marketing funnel and the flywheel covers which of the two planning models should lead.
Loop types
Most businesses can run more than one loop, but one is usually dominant, and knowing which decides where the experiments go.
| Loop type | What the customer produces | What turns it | What it needs to compound | Where it leaks most |
|---|---|---|---|---|
| Viral (invitation) | Invitations to people they know | A prompt at a moment of value; a reward or a reason to invite | A product that is better with others in it, or worth talking about | Few customers reach the moment of value; invitations go unaccepted |
| Content (user-generated) | Content, reviews, profiles, answers that search engines and feeds surface | Indexing and distribution of that content to new people | Content that answers what strangers search for; a path from content to sign-up | Content that nobody searches for; weak path from reader to customer |
| Content (company-produced) | Attention that funds more content | Distribution through search, social and email | Compounding authority and an audience that returns | Publishing without distribution; content detached from the product |
| Paid | Margin that funds the next acquisition round | Reinvesting contribution margin into paid channels | Payback faster than cash constraints allow; stable acquisition costs | Rising cost per customer as audiences saturate; margin assumed rather than measured |
| Sales (outbound) | References, case studies, expansion revenue | Sales capacity funded by expansion; referrals into similar accounts | Customers who succeed visibly and will say so | Long cycles; references unasked for |
| Marketplace (two-sided) | Supply that attracts demand, or demand that attracts supply | Matching; each side's presence drawing the other | Enough liquidity that both sides find what they came for | Imbalance: one side arrives and finds nothing |
| Product-embedded | A visible artefact carrying the product's name (a badge, a shared link, a document) | Recipients encountering the product in use | Frequent sharing as a natural part of using the product | Artefacts that hide the product; recipients with no reason to act |
The company-produced content loop is the one most businesses already run without calling it a loop, and its distribution half is where it usually breaks; the content distribution guide covers that half. The user-generated version depends on content that strangers search for and a programme for collecting it, which the user-generated content guide covers from the social side. The paid loop is the one most often mistaken for growth when it is only spending: it compounds only if the margin is real and the payback is faster than the cash runs out, which is the subject of the LTV and CAC guide.
Mapping a loop and its leaks
A loop is mapped on paper before it is measured in a warehouse. The method has four moves.
Draw the stages as a circle. Start with the acquisition event (a visit, an install, a first contact) and follow one customer round: what they do to get value, what they produce as a result, how that output reaches a stranger, and how the stranger becomes the next acquisition event. Each arrow is a stage. If the circle does not close, the business has a funnel with a referral afterthought, which is worth knowing.
Put a rate on every arrow. Each stage is a conversion with a measurable rate: the share of acquired customers who reach the value moment, the share of those who produce output, the share of output that reaches a new person, the share of new people who convert. Some rates will be guesses at first. Mark them as guesses.
Multiply round the loop. The product of the rates, together with the number of outputs each customer produces, gives the number of new customers each existing customer generates per cycle. If that figure is above one the loop compounds on its own; below one it decays but still subsidises other acquisition. Most real loops are well below one, which is normal: a loop that produces a fraction of a new customer per customer still cuts blended acquisition cost (CACCost to acquire one new customer.) substantially over time. Why it does, as the customers a loop produces go on to produce more and arrive cycle after cycle, is worked through in how a growth loop compounds.
Find the leak that matters. The leak is not the lowest rate but the stage where a feasible improvement moves the loop product most. A stage at a low rate that could plausibly double is a better target than one at a middling rate that is near its ceiling. That judgement is what the experiment backlog is built from, and the experimentation cadence guide describes how the backlog is prioritised and worked.
Mapping a design tool's invitation loop
A collaborative design tool maps its loop in five stages. Of 10,000 new sign-ups in a month, 4,200 create a first design (a 42% activation rate). Of those, 1,470 share a design with someone outside the account (35% of activators). Each sharer sends on average 2.3 share links, giving 3,381 recipients. Of recipients, 1,150 open the link and view the design (34%), and 310 of viewers sign up (27%). The loop therefore produces 310 new sign-ups from 10,000, a loop factor of 0.031. That looks small, but the recipients are free: blended CAC falls from £41 to £39.80 once the loop's sign-ups are included, and the team's model shows that lifting activation from 42% to 55% (the onboarding experiments already in the backlog) would raise the loop's output to over 400 sign-ups per 10,000 with no other change. The leak that matters is activation, not the share prompt, even though the share prompt is the stage the team had been arguing about.
Measuring throughput and cycle time
Once mapped, a loop is measured on two numbers that a funnel never needed. Throughput is how many new customers (or how much of the north star metric) the loop produces per period from a given input. Cycle time is how long one turn takes: from a customer's acquisition to the acquisition of the customer they produced. A loop with a strong factor and a long cycle may contribute less this year than a weaker loop that turns weekly, because compounding is a function of both.
| Measure | What it tells you | How it is read |
|---|---|---|
| Loop factor | New customers produced per existing customer per cycle | Product of stage rates and outputs per customer; above one compounds, below one subsidises |
| Throughput | Customers or north star units produced per period | Loop factor applied to the active base, tracked monthly by cohort |
| Cycle time | Time from one acquisition to the acquisition it causes | Median time between a customer's acquisition and their first produced customer |
| Stage rates | Where the loop leaks | Each arrow's conversion rate, by cohort, with its trend |
| Contribution to blended acquisition cost | What the loop is worth in money | Total acquisition spend divided by all customers including loop-produced ones, against spend divided by paid customers only |
Cohorts matter here because loop rates drift. A loop measured on the whole base mixes customers acquired last week with customers acquired two years ago who behave differently. The cohort analysis tutorial builds the table that makes stage rates readable by acquisition period, and the growth team's reporting KPIA metric tied directly to an objective. set should include the loop factor and cycle time alongside the north star, because a north star that rises while the loop factor falls means the growth is being bought rather than compounded.
One caution: loop measures are easy to flatter. Counting every customer who was "exposed" to a shared artefact as loop-produced, or attributing organic sign-ups to the content loop by default, inflates the factor and hides the paid spend that actually did the work. The attribution rule for loop-produced customers should be written down and applied by someone who does not own the loop. Marketplace loops add a second side to every measure: the guide to measurement for marketplaces covers a funnel per side, liquidity by area and category, and channel credit for supply judged by the demand it serves. The guide to growing a two-sided marketplace turns those measures into growth decisions: which side to recruit, where to start and when to open the next area.
Loops for non-software businesses
Loops are discussed mostly in software because software can instrument every stage. They exist elsewhere; they are just slower and partly dark.
Retail and consumer goods run a repeat-purchase loop (margin from the first order funds the acquisition of the next customer and the retention of this one) and, where products are visible or giftable, an exposure loop. The stages are measured through order data and reviews rather than product events, and cycle time is set by purchase frequency. A customer whose lifetime value (LTVTotal value a customer brings over time.) is realised over several orders is the loop's output, which is why repeat rate is the stage to map first.
Services and professional firms run a referral loop with a long cycle: a client who is well served tells a peer, who enquires months later. The leak is nearly always the ask. Most firms never prompt a referral at the moment of visible success, and measuring the loop means recording the source of every enquiry in a way the firm's CRM can report. Where the loop needs a formal scheme, the tutorial on designing a referral programme sets the reward from the margin and measures only the referred customers who would not have come anyway, and the referral reward test is the experiment design for deciding whether the referrer, the friend or both should be rewarded.
Local businesses run a review-and-search loop: served customers leave reviews, reviews lift local search visibility, visibility brings the next customer. It is slow, it is public, and it is one of the few loops where a single stage (asking for the review) is almost entirely within the business's control.
Content and media businesses run the company-produced content loop in its purest form: attention funds content, content earns distribution, distribution brings attention. The commercial stage, turning attention into subscriptions or advertising revenue, is where the economics are decided.
For all of these, the method is the same as for software: draw the circle, put a rate on each arrow, mark the guesses, and improve the stage where a feasible change moves the product most. The data is thinner, so the first job is often instrumenting a stage that has never been measured, and the first experiment is often simply asking.
Which loops a business can build depends on its motion, and the product-led versus sales-led growth comparison explains how each model shapes acquisition, activation and expansion.
For loops that must pass through an app store, the guide to growth loops for app installs counts store-page leakage and deep-link failures as separate arrows.
When growth flattens because every customer has to be bought, the growth plateau diagnosis helps confirm whether the missing loop, rather than saturation or churn, is the ceiling.
When the loop has to start from nothing at a product launch, growth loops for a product launch covers seeding a connected first group, a referral waitlist and reading the return by launch cohort.
Common mistakes
- Calling a funnel a loop. If the circle does not close with a measured arrow, it is a funnel. Say so and build the missing stage.
- Working on the share prompt before activation. Output comes from customers who got value; a loop with weak activation has nothing to share.
- Measuring the loop on the whole base. Rates drift by cohort; a loop read on blended data hides decay.
- Flattering the factor. Loose attribution of organic customers to the loop hides the paid spend doing the work.
- Ignoring cycle time. A loop that turns slowly may matter less this year than a weaker one that turns fast.
- Running the paid loop on revenue. Margin funds the next turn; revenue does not. Payback decides whether the loop can turn at all.
- Copying another company's loop. Loops depend on what customers produce naturally; a borrowed mechanism without that output is a feature nobody uses.
