Growth Loops: How to Tell If Your Product Actually Has One

Most "growth loops" teams draw on a whiteboard are not loops. Someone takes a funnel, curves the arrow back to the start, labels it a flywheel, and the team spends a quarter optimizing a mechanism that never actually fed itself. The diagram looked circular. The growth stayed linear.

This guide is about the decision underneath the drawing. It will help you decide whether your product has a real self-reinforcing mechanism, name the thing that actually cycles, separate the input that seeds the loop from the output it produces, prove how that output becomes next-cycle input, and locate the constraint that limits compounding, before anyone proposes a viral coefficient target.

Table of contents

• The direct answer: a loop is defined by re-entry
• A loop is a mechanism, not a metric
• What actually counts as a growth loop
• Funnel vs loop: two different questions
• The Loop Integrity Test
• Seed input vs loop output
• Choose the loop unit before you choose the metric
• The Growth Loop Contract
• The Output-to-Input Re-entry Map
• Growth loop mechanisms, not a fixed taxonomy
• Viral loop is one mechanism, not the whole category
• A referral program is not automatically a loop
• PLG is a motion; a loop is a mechanism
• Network effects and growth loops are different claims
• A flywheel diagram is not evidence
• Loop velocity is a constraint, not a vanity KPI
• Loop coefficient: the strict trust rule
• Quality-adjusted loop output
• Retention and activation are loop multipliers
• Saturation: loops end somewhere
• Loops depend on external channels too
• Loops can stack, but not into loop spaghetti
• Attribution is never perfectly clean
• Pattern is not cause
• Loop health is a system, not one number
• The Loop Constraint Diagnostic
• The Loop Review and Decision Record
• Worked case 1: a B2B collaboration loop
• Worked case 2: a public-artifact content loop
• Worked case 3: a two-sided marketplace loop
• Worked case 4: a paid funnel that is not a loop
• Worked case 5: a loop with bad quality
• Design the loop from value, not from the invite button
• Common failure modes
• The decision clinic: growth-loop questions teams actually bring
• Frequently asked questions
• References and further reading

The direct answer: a loop is defined by re-entry

A growth loop exists when activity in one cycle produces an output that becomes meaningful input for a future cycle. That is the whole test, and almost every failed loop design fails it.

The output must re-enter. Not "the process repeats." Not "we do this every week." The specific thing this cycle created has to become the thing the next cycle consumes:

• a shared document becomes the reason a new colleague joins;
• a published artifact becomes the reason a new user discovers the product;
• a completed transaction becomes the reason a new seller lists inventory;
• a retained account becomes the reason a second team inside the company adopts the product.

If there is no defensible arrow from this cycle's output back to next cycle's input, you do not have a loop yet. You may have one of several useful things, a funnel, a channel, a workflow, a campaign, a one-time referral event, but you should not call it a growth loop, and you should not measure it like one.

That single distinction prevents most of the damage this topic causes: teams chasing a viral coefficient for something that was never viral, leaders expecting compounding from a mechanism that was always linear, and roadmaps justified by a circular diagram nobody checked.

A loop is a mechanism, not a metric

The word "loop" describes how growth is produced. It is not a synonym for "compounding growth numbers," and it is not a metric. A loop is a named mechanism with a physical shape:

activity → output → exposure → new eligible input → (activation) → activity again

You can only measure a loop after you can describe that shape in your own product with named events and named people. If you cannot yet say what the output is, who receives it, and why that makes a new participant eligible, there is nothing to instrument, and a coefficient computed at that point is measuring your imagination, not your product.

This is why the field is so far ahead of most teams on definition and so far behind on diagnosis. Practitioner frameworks agree on the idea of a compounding system (see References); very few give you a way to localize which part of your specific mechanism is broken. That localization is the job of this page.

What actually counts as a growth loop

Here is the working definition this guide uses. It is deliberately strict, because loose definitions are why the term has stopped meaning anything.

A growth loop is a repeatable mechanism in which activity in one cycle produces an output that becomes meaningful input for a future cycle, generating additional adoption, usage, supply, demand, content, distribution or revenue.

Every clause is doing work.

• Repeatable mechanism, the cycle can run again without you manually rebuilding the entire acquisition effort each time.
• Output, the cycle produces something concrete: an invite that lands, an artifact someone can see, a listing, a review, a transaction, data, revenue, reputation, or audience exposure.
• Meaningful input, the next cycle consumes something of real value, not a vanity signup or junk invite.
• Future cycle, there is a genuine time gap; the input is used later, not within the same flow.
• Generating additional value, the mechanism produces more of the thing the business needs, not merely the same amount with more effort.

Notice what is not in the definition: exponential growth, a coefficient above one, virality, a Product-led motion, or a fixed number of steps. Those are consequences or subtypes, never requirements.

What a loop is not

Funnel vs loop: two different questions

A funnel and a loop are not competing models of the same thing. They answer different questions, and a healthy product usually needs both.

A funnel answers: Where do eligible users or accounts progress, and where do they drop out?

Visit → Sign up → Set up → Reach first value → Repeat

Funnels are about ordered progression and leakage. They are the right tool for diagnosing where eligible people fail to advance. (The full method, unit, eligible population, semantic steps, windows and drop-off diagnosis, belongs to the funnel analysis guide, not here.)

A loop answers: How does the output of one value-producing cycle create input for the next cycle?

Create useful artifact → Share it → New participant encounters value → New participant creates an artifact → ...

Loops are about feedback and compounding. They are the right tool for understanding whether growth feeds itself.

A loop usually contains funnels. The path from "new participant encounters the output" to "new participant becomes a loop-capable participant" is itself a funnel with drop-off. So when someone asks "where between invite sent and collaborator active are people dropping?" that is a funnel question, and it routes to the funnel owner. When someone asks "does collaborator activity create future participants?" that is a system question, and it is what this page owns.

Do not publish or believe "funnels are dead." Linear diagnosis and compounding mechanisms coexist. The expensive mistake is reading a funnel as if it were a loop, or defending a loop diagram with funnel metrics.

The Loop Integrity Test

The Loop Integrity Test is a gate, not a score. Walk it in order against one specific mechanism. The first question you cannot answer honestly is your actual constraint.

1. Value-producing action

What does a participant do that creates genuine value for themselves? If the action creates value only for you, it is marketing wearing a product costume.

2. Output

What does that action concretely produce? Possible outputs: an invitation, an artifact, published content, inventory, a review, a transaction, data, revenue, reputation, audience exposure.

3. Exposure

Who or what actually receives the output? Name the recipient. "The internet" is not a recipient.

4. Re-entry

How does that received output create new eligible input for the next cycle? This is the load-bearing question. If you cannot describe the arrow, stop and call it what it is.

5. New value

Does the next participant receive real product value, not just a notification? If the recipient gets nothing until they do repeated work, the loop is leaking at exposure.

6. Repeatability

Can the mechanism run again without you manually rebuilding the whole acquisition effort? If you must re-run a campaign each time, you have a channel with a manual effort step.

7. Quality

Does the next-cycle input have enough fit to matter? Junk invites, low-intent traffic, and fake accounts are inputs that poison the loop.

8. Participation

Can newly acquired or activated participants themselves perform the value-producing action? A loop where only founders can trigger the action is not self-sustaining. (Whether they do repeat value is a retention question; whether they ever reach first value is an activation question.)

The possible outcomes

No score, no percentage. A gate that produces a decision, not a number.

Seed input vs loop output

Almost every real loop needs an initial push. Pretending a loop "grows from nothing" is how paid spend gets mislabeled as a loop.

Three things must stay separate:

• Seed input, external input that starts or accelerates the mechanism: paid acquisition, SEO traffic, outbound sales, founder outreach, an existing user base, initial marketplace supply, a launch campaign.
• Loop output, what the current cycle creates: invites that land, artifacts published, listings added, transactions completed.
• Re-entry, how loop output becomes future input.

This distinction does real work. A product can be growing with a loop coefficient below one because a seed channel keeps refilling the top. That is not a failure; it is a business with a seed plus a partial loop, and the honest question is whether the loop you have is strong enough to matter relative to the seed cost. Conversely, a product can show a healthy coefficient because the seed is being counted as if it were loop output, a measurement error that hides the truth.

The mistake to avoid is subtle: paid acquisition is not the loop just because it keeps the cycle turning. Spend is seed. If revenue from acquired customers is genuinely reinvested into acquisition through a repeatable mechanism, that is a different claim with its own economics, and it is a reinvestment loop only when the reinvestment path is explicit, not assumed.

Choose the loop unit before you choose the metric

"The user" is the default unit in almost every growth article, and it is wrong for most products.

The loop unit is the thing that cycles, the entity that performs the value-producing action, produces the output, and comes back. Candidates: user, account, workspace, team, buyer, seller, creator, content item, transaction, listing, project, integration, dollar of revenue.

A real collaboration product is a good example of why one unit is not enough:

• The workspace triggers collaboration (the trigger unit).
• The invited user accepts the invitation (the acceptance unit).
• The workspace again is where retained value shows up.

If you compute "invites per user" you have silently chosen the user as the loop unit and you will misread your own mechanism. The invitation that matters was triggered by a workspace reaching a state where it needed another person, not by an individual user's propensity to share.

A marketplace makes the point sharper. If each buyer is assumed to independently bring new buyers, a one-unit coefficient is not merely imprecise; it is the wrong model. Buyer activity increases the value that attracts sellers, and seller activity increases the value that attracts buyers. That is a two-sided mechanism, and it needs a two-sided description before any number makes sense.

Choose the loop unit that matches the mechanism. Then choose a second unit only when the mechanism genuinely has two roles, like buyer and seller, or creator and viewer.

The Growth Loop Contract

Write the loop down before you build or optimize anything. This contract is copyable and deliberately excludes any health score or confidence percentage.

# Growth Loop Contract

Growth objective:
[What business/Product outcome is this loop meant to produce?]

Loop unit:
[User / Account / Workspace / Team / Buyer / Seller / Creator /
 Content item / Transaction / Listing / Other, and why this unit]

Seed input:
[What external or one-time input starts or accelerates the cycle?]

Value-producing action:
[What a participant does that creates real value for themselves]

Output created:
[What the action concretely produces]

Who/what receives the output:
[The named recipient of the output]

Re-entry mechanism:
[How the output becomes new eligible input for the next cycle]

Next-cycle input:
[What the next cycle actually consumes]

Activation needed before a new participant can trigger the loop:
[What first value / qualified state must occur]

Quality condition:
[What makes the next-cycle input "good enough" to matter]

Retention / repeat condition:
[What makes a participant able to run the cycle again]

Primary loop metric:
[One metric, with unit, numerator, denominator]

Supporting metrics:
- [...]

Loop delay / latency:
[Time from output to usable next-cycle input, with its unit]

Known leakage points:
- [...]

Saturation constraint:
[What finite resource, audience or channel limits repetition?]

External dependency / channel dependency:
[Which external platform, channel or team the loop relies on]

Instrumentation source:
[Where the loop can actually be observed]

Current strongest uncertainty:
[The thing you are least sure about]

Next evidence / intervention:
[The next method or change that matches the constraint]

Version / owner / date:
[...]

The contract is useful precisely because it forces the re-entry sentence to exist. A team that cannot fill in "re-entry mechanism" has just discovered it has a funnel.

The Output-to-Input Re-entry Map

The contract names the parts. The re-entry map shows the arrows, and it is where a circular drawing either earns its shape or collapses.

Seed input
   │
   ▼
Value-producing action
   │
   ▼
Output
   │
   ▼
Distribution / exposure
   │
   ▼
Qualified next-cycle input
   │
   ▼
Activation into a loop-capable participant
   │
   ▼
(re-entry) ───────────────► back to Value-producing action

For each node in the map, record six things:

• owner, who is accountable for this step;
• metric, the observable event that marks it;
• delay, the typical time to traverse it;
• friction, what makes it harder than it should be;
• quality condition, what makes the output usable downstream;
• evidence, how you actually know it happened.

If any arrow cannot be drawn with an owner, a metric and a plausible delay, that is your constraint, and it is almost always more useful than a coefficient.

The map is intentionally text-first. It stays readable on a phone, works with a screen reader, and cannot be mistaken for decoration. A circular SVG with eight tiny labels is not a better artifact; it is a worse one that hides the missing arrow.

Growth loop mechanisms, not a fixed taxonomy

Any list of loop "types" should be treated as a vocabulary, not a law. The classic public framing used viral, content, paid and data loops; that set is a useful starting point but is incomplete and can mislead if treated as the full menu. Structuring loops by value-producing mechanism is more honest:

Two rules for using this table. First, do not force symmetry, these mechanisms are not equally common, equally powerful or equally available to a given product. Second, do not treat the presence of a mechanism class as proof of a loop, the last column is the real test. A referral feature is an invitation mechanism, not automatically a referral loop. A public artifact feature is a content mechanism, not automatically a content loop.

Viral loop is one mechanism, not the whole category

A viral loop is a loop in which participants cause the acquisition of new participants through sharing or invitation, fast enough and reliably enough that it materially contributes to growth. It is one mechanism class among many.

Growth loops can be built on content, marketplace liquidity, account expansion, creator output, reinvestment, data improvement or organic distribution. Collapsing "growth loop" into "viral loop" causes a specific, expensive failure: teams without natural virality conclude they have no loop, and teams with natural virality conclude that invites are the only loop worth building.

A useful test: if your product has no natural reason for one participant to expose it to another, the honest answer may be that you have a content, acquisition or sales loop, or no loop at all, and forcing an invite mechanism will produce spam, not compounding.

A referral program is not automatically a loop

A referral CTA is an invitation event. Whether it is a loop depends on answers to five questions:

1. Why does the referrer act? Genuine product value, or only an incentive?
2. What value does the referred person receive? Real value, or a discount that expires with the interest?
3. Does the referred participant become a future referrer? If not, the mechanism is one step deep and does not compound.
4. Is the mechanism repeatable? Can it run without a fresh campaign each time?
5. Does quality hold, and is the cost sustainable? Fraud, low-fit signups and incentive dependence all break the economics.

If the honest answers describe a one-time introduction, a referral program is a tactic and a channel, not a loop. That is not a criticism; it is a routing decision. Channel selection and referral tactics belong to the acquisition and PLG owners. This page owns the question of whether the referral mechanism actually cycles.

PLG is a motion; a loop is a mechanism

These get conflated constantly. They are different kinds of claims.

• Product-led growth describes a go-to-market and growth motion: prospects experience value before human sales intervention, and self-serve acquisition, activation and expansion carry much of the journey.
• A growth loop describes a repeating mechanism: this cycle's output becomes next cycle's input.

A PLG motion may contain one or more loops (a collaboration loop, a public-artifact loop, self-serve expansion). A PLG motion can also exist with no meaningful self-reinforcing loop, plenty of PLG products grow through a strong free tier, great onboarding and efficient paid acquisition, with nothing that compounds on its own. And a loop can exist in a business that is not PLG at all: sales-assisted account expansion and marketplace liquidity both produce real loops without self-serve growth.

Teach the difference plainly: PLG says how users get and grow value; a loop says how one cycle feeds the next. If the user's question is how signup, self-serve onboarding, a free experience, product-qualified leads and expansion should work together, that is the PLG owner's job.

Network effects and growth loops are different claims

Network effects and growth loops are routinely used as synonyms. They are not.

• A network effect describes how product value changes as participation changes. More participants make the product more valuable to each participant.
• A growth loop describes how a repeating mechanism creates future input, new users, supply, content, revenue.

They often reinforce each other, and a strong network effect is a reason a loop can be durable. But one does not imply the other:

• A product can have a network effect and no acquisition loop. Value may rise with participation while nothing in the product generates new participants.
• A product can have an acquisition loop and no network effect. A referral incentive can bring users in without making the product more valuable as it grows.

Network-effect research is its own large literature (see References); this page only needs the boundary. Do not call every circular mechanism a network effect, and do not assume a network effect will automatically produce a loop.

A flywheel diagram is not evidence

"Flywheel" is used loosely to mean anything circular and virtuous. A circular strategy diagram is not a growth loop. To earn the word, a flywheel needs the same four things any loop needs:

• a real output;
• re-entry into future input;
• an observable mechanism;
• an observable cycle, you can see it happen more than once.

If the flywheel is a slide, not a mechanism, call it a strategy narrative and move on. The vocabulary matters because teams make budget decisions based on which word is used: "flywheel" often implies "leave it alone, it compounds," while "loop with an unproven re-entry arrow" implies "go instrument it."

Loop velocity is a constraint, not a vanity KPI

Loop velocity is the time between generating loop output and creating usable next-cycle input. Depending on the product, that interval might be:

• time from invitation sent to collaborator active;
• time from artifact published to a new creator discovered;
• time from listing created to matched transaction;
• time from purchase to the buyer becoming a repeat participant;
• time from a retained account to the next team adopting.

It is a real and often under-appreciated constraint. A fast loop compounds more times per year than a slow one, all else equal. But "faster is always better" is false, and treating velocity as an unconditional KPI produces real damage:

• an enterprise collaboration loop that takes weeks can be genuinely high-value; forcing it faster may mean skipping approvals or security review;
• marketplace supply may need quality review, and removing that review to speed the cycle degrades the very liquidity the loop depends on;
• rush-inviting from prompts raises volume while lowering invitation fit and trust;
• shortening onboarding to speed re-entry may strip the step that creates retained value.

So treat delay as a system constraint to understand, not a number to minimize. The useful question is not "how do we make the cycle faster?" but "which delay is binding on our growth objective, and what does removing it cost?" That question belongs in the constraint diagnostic below, and any resulting intervention belongs in an experiment plan, not straight into the product.

Loop coefficient: the strict trust rule

A "loop coefficient", sometimes a viral coefficient or K-factor, is a ratio of new participants produced per participant. The naive version, new users ÷ users who triggered the loop, is unusable on its own because it hides every decision that makes it meaningful.

Before you trust any loop coefficient, specify:

• unit, new what per which unit? (users per workspace, buyers per seller, not "users per user" by reflex)
• time window, over what period? A coefficient without a window is not a rate.
• eligible population, who counted as able to produce output?
• quality criteria, did the new participant activate, or merely sign up?
• re-entry qualification, did the new participant become capable of producing the next output?
• direct vs multi-step, is this one step of sharing, or a multi-step chain? Mixing them inflates the number.

Only then is a coefficient interpretable. And even then:

• Do not publish a universal "above 1 is good" rule. A coefficient above one can describe a product that is flooding itself with low-quality participants. A coefficient below one is normal and can describe a healthy business that pairs a modest loop with a seed channel.
• A coefficient is not a target. Once it is a target, the cheapest way to move it is usually to degrade quality, more invites, looser fit, incentives that attract arbitrage. The metric improves while the product worsens.

A coefficient below one is not failure. Inspect the unit, the seed, the economics and the product context first. A coefficient above one is not proof of health. Inspect quality and retention first.

Quality-adjusted loop output

Raw signups exaggerate loop health. The output a loop should be judged on is the output that actually matters downstream:

• an activated referred user, not a signup;
• a retained workspace, not a created one;
• a valid buyer–seller match, not a listing;
• a qualified lead, not a form fill;
• a completed transaction, not a cart;
• a viewer who reaches value, not a pageview.

The principle is simple: the loop output should match the next-cycle input that actually matters. If your loop's next-cycle input is a creator who will publish, then "new signups" is the wrong output metric, "new signups who publish within 14 days" is the right one.

This is where most "successful" loops quietly fail. The mechanism runs, the top-line number rises, and the quality of what enters the next cycle falls. Over time the loop gets louder and weaker at once.

Retention and activation are loop multipliers

A loop can produce new participants quickly and still fail, because the new participants never become real loop members:

• they never reach first value → activation problem;
• they reach value once but never repeat → retention problem;
• they activate and retain but never become able to produce output → re-entry problem;
• they participate in volume but degrade network quality → quality problem.

The handoff matters more than the diagnosis here. If loop output is entering the product but not becoming retained value, the answer is not more amplification. More amplification into a leaking product accelerates the leak. The right next step is to diagnose activation or retention, using the activation guide when participants never reach first value, or the retention system when they reach it but do not come back.

This page owns the loop's mechanism. It does not own the definition of repeated value, and it should not try to solve retention by turning up the volume on acquisition.

Saturation: loops end somewhere

Loops do not compound indefinitely. Every loop has a finite resource, and naming it is part of the design:

• a finite personal or professional network (invite exhaustion);
• channel saturation (the audience has seen the tactic);
• SEO competition (the content surface fills up);
• content supply saturation (more artifacts than demand);
• marketplace side imbalance (supply outruns demand or vice versa);
• referral fatigue and incentive arbitrage;
• geographic or regulatory limits;
• enterprise seat caps and procurement cycles.

Exponential language without a saturation term is a modeling error, not optimism. A loop that looks healthy on a short window may simply be early. The useful question is "what is the finite resource this loop consumes, and how much of it is left?", and the answer belongs next to the coefficient, not in a footnote.

This is also why "our loop will compound forever" and "our loop is failing" can both be wrong: the same mechanism can be strong early, saturate later, and need a transition to a new loop or channel. Practitioner frameworks describe exactly this pattern, a product is typically powered by one or two major loops that transition over time, not an ever-growing collection.

Loops depend on external channels too

Even a "product-led" loop usually leans on something external: Google or an app store for distribution, a social platform for exposure, paid traffic to seed, a sales team for enterprise expansion, or a creator community for supply. These dependencies are not embarrassments to hide; they are failure risks to map.

For each loop, record:

• the internal mechanism, what your product and users do;
• the external dependency, the platform, channel or team it relies on;
• the failure risk, what happens if that dependency changes its rules, prices or access.

This is the practical form of product–channel fit: channels define the rules, and products must be molded to fit them. A loop that depends entirely on one platform's API is one policy change away from stalling, and that risk belongs in the contract, not in a surprise postmortem.

Loops can stack, but not into loop spaghetti

Loops can feed each other. A content loop can seed acquisition, which feeds a collaboration loop, which drives account expansion. Stacked loops are real, and mature products often run more than one.

The failure mode is building the stack before understanding the parts. A diagram with seven loops and twelve arrows is not a strategy; it is a way to hide the binding constraint. When everything connects to everything, no one can tell which loop is actually limiting growth.

So: map one loop clearly first. Prove its re-entry arrow, find its constraint, measure its quality. Only then map the interfaces between loops, where one loop's output becomes another's seed, and only then decide whether to invest in a second loop. There is no universal "one loop first" timing rule, but there is a universal diagnostic cost to confusing yourself.

Attribution is never perfectly clean

Loop-driven acquisition is often hard to attribute, and pretending otherwise is its own failure mode. Consider:

• a public artifact seen in a feed, followed by a direct signup days later;
• an invitation plus a branded search that lands the same person;
• a multi-device journey with no shared identifier;
• a marketplace match where the network, not any channel, produced the buyer;
• offline sharing of a link, a document or a recommendation.

The honest model separates three things:

• direct attribution, you can observe the path with reasonable confidence;
• inferred contribution, the evidence suggests the loop contributed, but the exact path is uncertain;
• unattributed / unknown, you cannot say, and should not pretend you can.

Do not force every loop into UTM parameters. A loop that is real but under-instrumented is a measurement problem to acknowledge, not a reason to declare the loop dead or to overstate it. The detail of how to instrument it, event semantics, identity, source of truth, QA, belongs to the measurement contract guide; this page owns the honest boundary between what the loop produces and what you can actually observe.

Pattern is not cause

A localized pattern in loop data is not a cause. Two recurring examples:

Bad inference: "Invite acceptance fell, therefore the invite copy is bad."

Competing explanations: a lower-fit inviter cohort, lower-fit invitees, a changed role mix, a permissions or email-deliverability issue, spam fatigue, an instrumentation break, or a genuinely slower decision cycle.

Bad inference: "Loop velocity improved, therefore growth is healthier."

Competing explanations: a change in what counts as output, a quality drop that made the cycle faster in name only, a segment mix shift, or a shorter measurement window.

The discipline is the same one that applies to any funnel or cohort pattern: generate competing explanations first, then find the evidence that would discriminate between them. Only a discriminating result justifies an intervention. This is also why loop diagnosis should not reflexively become an A/B test, sometimes the next method is instrumentation, segmentation, or qualitative research, and forcing a test too early wastes traffic and confidence.

Loop health is a system, not one number

There is no universal loop-health score, and you should distrust any that claims to be one. A loop is healthy or unhealthy across several dimensions at once:

• trigger frequency, how often the value-producing action occurs;
• output quality, how good the thing produced actually is;
• reach, how much relevant exposure the output gets;
• next-input creation, how often output becomes qualified input;
• activation, whether new participants reach value;
• time-to-re-entry, how long the cycle takes;
• retention/quality, whether participants and network quality hold;
• saturation, how much of the finite resource remains.

The reason to resist a single number is that a loop can improve one component and worsen the product overall. Making invites easier can raise trigger frequency while lowering output quality. Speeding the cycle can raise velocity while cutting activation. Compressing everything into a score hides exactly the trade-offs you need to see.

(If you need a headline metric for a loop, define one, but define it with unit, numerator and denominator, and keep the system view next to it.)

The Loop Constraint Diagnostic

This is the decision system for localizing the binding constraint. Work top to bottom; stop at the first stage where the evidence shows a real break. For each stage: what the evidence would look like, what it does not prove, and where the next method lives.

The rows are not ranked by importance; they are ranked by the order in which localizing beats guessing. The most common diagnostic error is stopping at a downstream row (say, "conversion is low") when the real break is upstream (say, "the output reaching recipients is low quality").

The Loop Review and Decision Record

When you have applied the diagnostic, record the decision. This artifact is copyable and deliberately contains no confidence percentage.

# Growth Loop Review

Loop hypothesis:
[...]

Loop unit:
[...]

Seed input:
[...]

Value-producing action:
[...]

Output:
[...]

Re-entry mechanism:
[...]

Qualified next-cycle input:
[...]

Activation / retention requirement:
[...]

Current evidence:
- [...]

Strongest constraint:
[Trigger / Output quality / Reach / Conversion / Activation /
 Re-entry / Delay / Retention / Saturation / Measurement / Unknown]

Competing explanations:
1. [...]
2. [...]
3. [...]

Next evidence method:
[...]

Proposed intervention:
[...]

Primary metric:
[unit, numerator, denominator]

Guardrails:
- [...]

Decision:
[Investigate / Instrument / Test / Ship / Hold / Abandon loop hypothesis / Monitor]

Revisit trigger:
[What event or date forces a re-review]

Owner / date:
[...]

No confidence score. The value of the record is that it preserves the reasoning, especially the competing explanations, so the next review does not restart from the diagram.

Worked case 1: a B2B collaboration loop

Setup (hypothetical). A team research workspace. A researcher builds an evidence board for a decision, then needs a teammate to review it before sharing with stakeholders.

Why the invite itself is not the loop. The invitation is one event. If the invited teammate reviews once and disappears, the mechanism is one step deep, an invitation mechanism, not a loop. The loop only exists if the invitation produces a teammate who then creates or joins more work.

The contract, filled in.

• Loop unit: workspace for the trigger; user for acceptance; workspace again for retained value.
• Seed input: founder-led outbound and a small paid pilot that fills the first workspaces.
• Value-producing action: a researcher creates a board that is useful enough to need a second opinion.
• Output: a specific invitation to review a specific board, the output is the shared board plus the reason to look at it, not the email.
• Re-entry mechanism: the invited teammate, having received real value from the board, creates their own board or invites a third person into the next decision.
• Activation requirement: the teammate must open the board and take one meaningful action (comment, edit, add evidence), not merely accept.
• Quality condition: the board is substantive enough that reviewing it is genuinely useful.
• Delay: typically days for a first review; weeks for the teammate to become an independent creator.

Where it leaks, and what the evidence shows. The honest failure story: invitations are sent and accepted quickly, but only a small share of invited teammates create or join further work. The loop is blocked at re-entry. A naive reading says "increase invites." The correct reading is that invited teammates are not activating into loop-capable participants, they accept, glance, and leave.

The next decision. Route to activation: define what first value means for a collaborator (not for an owner) and instrument it. Do not turn up invitation prompts; that would raise trigger frequency while worsening output quality and trust. If the instrumented data later shows collaborators do activate but do not return, the routing changes to retention.

What this case teaches. The invite is the visible event; re-entry is the real mechanism. The constraint was re-entry/activation, not amplification, and the cheapest wrong move (more invites) was also the most tempting.

Worked case 2: a public-artifact content loop

Setup (hypothetical). A portfolio/report creator tool. Users build a designed artifact and can publish it at a public link.

The mechanism.

User creates a useful public artifact
→ publishes/shares it
→ a viewer gets real value from the artifact
→ the viewer adopts the product
→ the viewer creates a new artifact
→ (re-entry) that artifact gets discovered by more viewers

The contract, filled in.

• Loop unit: the content item for exposure; the creator for re-entry.
• Seed input: existing users plus a launch surface (a gallery, a directory, a social push).
• Value-producing action: create an artifact that is genuinely useful or impressive to a viewer.
• Output: the published artifact and its attribution.
• Re-entry mechanism: viewers who reach value become creators, producing new discoverable artifacts.
• Activation requirement: the viewer must not just see the artifact but reach the point where they want to make one.
• Delay: long, from publication to discovery can be weeks, especially if discovery depends on search.

Where it can fail, and the failure modes are many:

• artifacts are never actually public (default private, or sharing is buried);
• viewers see the artifact but do not perceive the product as the reason it looks good;
• attribution is weak (the artifact is screenshotted or exported without a link);
• creators remove or pay to remove branding;
• content quality is low, so discovery surfaces do not reward it;
• viewers arrive but do not activate, the loop is blocked at qualified input;
• discovery is saturated: more artifacts than there is demand to view them.

The next decision. If viewers arrive but rarely activate, the constraint is conversion/activation, not publication volume. Publishing more artifacts into a non-converting loop multiplies the weakest part. Instrument the viewer→creator path before adding publishing prompts.

What this case teaches. A content loop is not proven by traffic. It is proven when viewers become creators who publish. Attribution weakness is common and should be stated honestly, not papered over with a high traffic number.

Worked case 3: a two-sided marketplace loop

Setup (hypothetical). A local services marketplace with sellers (providers) and buyers (customers), where liquidity is geographic.

The mechanism.

A seller lists quality supply
→ buyers in that area get a better experience
→ more buyers transact
→ sellers see more demand
→ more sellers list quality supply
→ (re-entry) better supply, more demand

The contract, filled in.

• Loop unit: supply side (seller) and demand side (buyer) as two roles, a single user coefficient is the wrong model.
• Seed input: supply seeding in a single city, often manually recruited, plus local demand marketing.
• Value-producing action: a seller lists genuinely available, quality supply.
• Output: a match (a listing that produces a completed transaction).
• Re-entry mechanism: completed demand gives sellers a reason to list more, and more supply improves the buyer experience.
• Delay: can be long on the supply side and very different per side.

Where it can fail:

• supply grows but buyers do not transact → cross-side constraint, not "more supply = stronger loop";
• quality falls as supply scales (fake listings, poor fulfillment), degrading the buyer experience;
• liquidity is local, national supply numbers hide a dead city;
• one side grows faster and matching friction rises;
• same-side competition (sellers competing) creates negative direct effects that can outweigh the indirect benefit;
• the value curve is asymptotic, early supply adds a lot, later supply adds little (wait times barely improve past a threshold).

The next decision. If supply grows but liquidity does not, diagnose the cross-side mechanism in the affected geography and quality segment, and instrument the match and completion, not the listing count. Do not reduce a marketplace loop to a single referral coefficient; it will hide exactly the side that is broken.

What this case teaches. Two-sided loops need a two-sided model and a local view. The most common error is reading supply growth as loop health when the demand side never responded.

Worked case 4: a paid funnel that is not a loop

Setup. Paid ad → landing page → signup → purchase.

This is a funnel fed by a channel. It is not a loop, for a precise reason: nothing purchased re-enters as future acquisition input. The ad spend is seed, and if you stop spending, the flow stops.

That does not make it bad. It makes it a linear channel with a cost, and the correct questions are acquisition-economics questions, cost per acquisition, payback, margin, retention of acquired customers, which belong to the acquisition and monetization owners.

When it becomes a loop. Only when there is a real, repeatable reinvestment mechanism:

Revenue / margin
→ reinvested into acquisition with known CAC and payback
→ new customers
→ revenue / margin
→ reinvest again

Even then, this is constrained by CAC, payback period, margin, channel saturation, retention and cash timing. It is not "paid acquisition compounds automatically." A reinvestment loop is a claim about economics and re-entry path, and it must be stated as one, with the seed (spend) clearly separated from the loop output (retained revenue) and the re-entry (reinvestment rule).

Why the classification matters. If a team calls this a loop, it will expect compounding and underinvest in the channel economics. If it calls it a channel, it will manage payback honestly. The name changes the decision.

Worked case 5: a loop with bad quality

Setup. A referral incentive drives a surge of signups.

The pattern. Raw output looks excellent: signups rise sharply. But downstream:

• activation is low, many signups never reach first value;
• some portion is fraudulent or self-referential (people gaming the reward);
• churn is high, the acquired users were attracted by the incentive, not the product;
• support burden rises as confused users ask about rewards;
• margin erodes as the incentive cost outruns the value of the acquired users.

The correct reading: more loop output ≠ healthier growth. The loop coefficient improved and the product got worse. The output that entered the next cycle was low quality, so the next cycle's participants were also low quality, and the mechanism degrades itself over time.

The fix is not "add a bigger incentive." It is to re-specify the quality condition, the output should be activated, retained participants, not signups, add fraud guardrails, and reconsider whether the incentive is compensating for weak product value. If the incentive is structurally what drives participation, the loop may be an incentive-funding mechanism rather than a product loop, and its economics should be judged as such.

What this case teaches. Quality-adjusted output is the only honest output. A coefficient computed on junk is not a growth signal; it is a warning.

Design the loop from value, not from the invite button

The most common way to build a fake loop is to start with the sharing mechanic:

"How can we make users invite?"

The better starting question is:

"What value-producing action naturally creates something useful to another participant, or creates future growth input?"

This ordering matters because it determines whether the exposure feels like a natural consequence of value or an interruption. Users detect and resist artificial sharing. Loops built on forced contact import, dark-pattern prompts and unnecessary friction produce short-term volume and long-term distrust, and, worse for the loop, they produce the low-quality input described in the previous case.

Practical implication: start from the value-producing action, then ask what output it naturally produces, then ask who would genuinely benefit from receiving it. Only then design the exposure. If no honest answer emerges, the product may not have a loop, and that is a finding, not a failure.

Common failure modes

Circle diagram theater. A circular drawing with no observable re-entry. The team optimizes a mechanism that never fed itself.

Referral = loop. A single invite event is mislabeled as a compounding mechanism. One step does not compound.

Funnel = loop. Sequential conversion stages are drawn with a curved arrow but no output→input feedback. Different question, different method.

Coefficient theater. One number ignores unit, time, eligible population and quality. It improves when the product worsens.

Velocity theater. A faster cycle is pursued at the cost of fit, trust or activation. Speed is not value.

Vanity signups. Output is measured as raw signups, so the loop fills itself with low-quality participants.

Retention blindness. The loop keeps filling the product with participants who do not stay. More amplification accelerates the leak.

Saturation blindness. A finite audience, network or channel is ignored, and a healthy-looking loop is just early.

Attribution certainty. Multi-touch and indirect exposure are narrated as exact. The model is cleaner than the world.

Forced sharing. Growth tactics harm user value, with the side effect of poisoning loop input quality.

Famous-company cargo cult. A mechanism is copied from a company without its context, economics or channel fit. Mechanism without context is decoration.

Too many loops. A seven-loop map hides the binding constraint. Complexity mistaken for sophistication.

Product-led dogma. External channel and sales dependencies are ignored because the loop is called "product-led." The dependency is still a failure risk.

The decision clinic: growth-loop questions teams actually bring

Frequently asked questions

What is a growth loop?

A growth loop is a repeatable mechanism in which activity in one cycle produces an output that becomes meaningful input for a future cycle, generating additional adoption, usage, supply, demand, content, distribution or revenue. The defining feature is re-entry: the output has to feed the next cycle. Without that, you have a funnel, a channel or a workflow, useful things, but not a loop.

What is the difference between a growth loop and a funnel?

A funnel answers "where do eligible users or accounts progress and drop out?" A loop answers "how does the output of one cycle create input for the next cycle?" Funnels diagnose linear progression and leakage; loops describe self-reinforcing mechanisms. They are complementary, not competing. A loop usually contains funnels, and a healthy product usually uses both.

What are examples of product growth loops?

Real examples are mechanisms, not brand names: a collaboration loop (shared work invites a needed collaborator who then invites others), a public-artifact loop (viewers become creators who publish discoverable artifacts), a marketplace loop (supply improves the buyer experience, which attracts demand, which attracts supply), and a sales-assisted expansion loop (a retained account advocates internally, adding teams). Famous-company versions of these exist, but the mechanism and its context matter far more than the logo.

What is a viral loop?

A viral loop is one mechanism class in which participants cause new participants through sharing or invitation, reliably enough to contribute materially to growth. It is a subtype of growth loop, not a synonym. Many growth loops are not viral at all, content, marketplace, reinvestment and account-expansion loops can all compound without virality.

Is a referral program a growth loop?

Not automatically. A referral program is an invitation mechanism. It becomes a loop only if the referred participant becomes a future referrer, quality holds, and the mechanism is repeatable. If it is a one-time introduction, it is a tactic and a channel, which belongs to the acquisition and PLG owners, not to loop design.

What is a growth-loop coefficient?

A coefficient measures new participants produced per participant, but it is only meaningful with six things specified: unit, time window, eligible population, quality criteria, re-entry qualification, and whether the output is direct or multi-step. Without those, it is not interpretable. The naive "new users ÷ users who triggered" hides every one of those decisions.

How do you measure a growth loop?

Measure the system, not one number. Instrument each node of the Output→Input Re-entry Map: trigger frequency, output quality, reach, qualified next-input creation, activation, time-to-re-entry, retention/quality and saturation. Every metric needs a unit, population, time window, definition, data source and quality condition. Detail of the measurement contract belongs to the instrumentation owner.

What is loop velocity?

Loop velocity is the time between generating loop output and creating usable next-cycle input, for example, from invitation sent to collaborator active, or from artifact published to new creator discovered. It is a real constraint, but "faster is always better" is false: speeding a cycle can mean skipping approvals, quality review or the steps that create retained value.

Does a loop need a coefficient above 1?

No. A coefficient below one is normal and can describe a healthy business that pairs a modest loop with a seed channel. A coefficient above one is not proof of health, it can describe a product flooding itself with low-quality participants. Inspect unit, seed, economics, quality and retention before judging.

Can B2B products have growth loops?

Yes, and often slow ones. Collaboration loops (shared work invites collaborators), sales-assisted expansion loops (retained accounts add teams), and content loops (reports, case studies, templates) all apply. B2B loops usually involve team decisions, approvals and longer delays, so velocity is a constraint to understand rather than minimize. The worst habit is to assume B2B cannot loop and to underinvest in the mechanism that already exists.

Can marketplaces have growth loops?

Yes, and they are usually two-sided. Supply improves the buyer experience, which attracts demand, which attracts supply. But cross-side effects, local liquidity, quality, same-side competition and an asymptotic value curve mean a single user-level coefficient will mislead you. Model both sides and check liquidity locally.

What is a content growth loop?

A content loop is a mechanism where creating and publishing useful content leads to discovery, which leads to new participants who create more content. It is only a loop if discovery produces new creators, not just traffic, and if attribution, discoverability, content quality, spam and SEO lag are handled honestly. Not every user-generated-content product has a content loop.

What is the difference between product-led growth and growth loops?

PLG describes a go-to-market and growth motion: prospects experience value before human sales intervention, with self-serve acquisition, activation and expansion. A growth loop describes a repeating mechanism where output becomes future input. A PLG motion may contain loops, may contain none, and loops can exist in non-PLG businesses. They are different kinds of statements.

Does every product need a growth loop?

No. Many products grow well through acquisition channels, sales, partnerships, community or paid media without a self-reinforcing loop. A loop is one system for producing growth, not a maturity badge. Forcing a loop onto a product with no natural re-entry usually produces spam, low-quality input and wasted roadmap, which is worse than an honest linear channel with managed economics.

How do you find the bottleneck in a growth loop?

Work the Loop Constraint Diagnostic in order: trigger, output quality, reach, conversion, activation, re-entry, delay, retention, saturation, measurement. Stop at the first stage where the evidence shows a real break. Then record competing explanations before choosing an intervention, because a localized pattern is not a cause. The bottleneck is usually upstream of where the pain first becomes visible.

References and further reading

• Brian Balfour, Casey Winters, Kevin Kwok and Andrew Chen, Growth Loops are the New Funnels (Reforge, 2018). The origin of the contemporary growth-loop framing: funnels are too micro to answer "how does the product grow"; loops are closed systems where inputs through a process generate more output that is reinvested in the input. Labels the funnel-vs-loop distinction, note that loops are typically one or two major mechanisms that transition over time, and warns against drawing many low-powered loops. Framing, not measurement law.
• Reforge, Growth Wins When Built On A Solid Foundation of Retention & Engagement and Building Growth: Product, Process, and Team (2018). The context series: compounding loops as a function of the existing user base, and the hypothesis-driven growth process. Practitioner framing.
• Brian Balfour, Product Channel Fit Will Make or Break Your Growth Strategy (2017). Products are molded to channels, not the reverse; distribution follows a power law; channel fit can break when a channel changes its rules. The basis for treating external dependencies as loop failure risks.
• NFX, The Network Effects Manual (2021). A structured taxonomy of network effects, and an explicit statement that network effects are not viral effects and have different objectives and playbooks; includes data network effects and the asymptotic marketplace value curve. Used here as the network-effects boundary, not as a growth-loop model.
• Wikipedia, K-factor (marketing). The standard k = i × c definition, its epidemiological origin, and the note that a k-factor below 1 can still accompany viral growth. Used only as a definitional reference; no benchmark or universal threshold is taken from it.

All quantitative claims about specific companies were deliberately not reproduced here. The sources above are used for framing, definitions and boundary distinctions; where this guide needed an example, it used a labelled hypothetical so the mechanism, not a logo, carries the lesson.