Growth Loops and Experimentation

Funnels leak; loops compound. A funnel pours users in the top and loses them out the sides, and when you stop pouring, growth stops. A loop uses the output of one cycle as the fuel for the next, so it keeps turning after you stop pushing. The best growth isn't a campaign you run; it's a machine you build into the product.

The concept

Two mental models for growth, and the difference matters enormously.

 FUNNEL          awareness → interest → signup → paid → (leaks out the sides)
                 linear · you refill the top forever · stops when you stop

 LOOP            action → output → that output brings the next user → repeat
                 circular · each cycle fuels the next · compounds on its own

A growth loop is a closed cycle where the output of using the product generates the input for acquiring or retaining the next user. It compounds because each turn feeds the next turn. Funnels are how you think about conversion; loops are how you build durable growth.

📐 Build loops, not campaigns, and sequence retention → loop → scale. A funnel stops the moment you stop pouring; a loop keeps turning because its output becomes its own next input. But a loop only compounds on top of working retention, stacked on a leaky product, it just makes the failure louder.

The main loop types

The best products stack more than one. But a loop only works on top of retention (Chapter 50), a viral loop that acquires users who immediately churn just accelerates the leak. Fix retention first; then a loop compounds it.

Why loops beat funnels

• Funnels are rented; loops are owned. A funnel needs constant refuelling (ads, campaigns). A loop, once built, keeps turning.
• Loops compound; funnels are linear. Double the input to a funnel, double the output. A loop's output becomes the next input, that's exponential, not linear.
• Loops are defensible. A network or content loop gets stronger with scale, which competitors can't easily copy.

You still use funnel thinking to find and fix conversion leaks within a loop. Loops and funnels aren't rivals, loops are the growth engine, funnels are the diagnostic lens.

Experimentation: how you actually improve growth

You don't guess your way to a working loop; you test. Disciplined experimentation is the method:

 1. OBSERVE     the data shows a problem or opportunity (Chapter 49)
 2. HYPOTHESISE "if we change X, metric Y improves, because Z"
 3. TEST        smallest experiment that gives a real signal
 4. MEASURE     against a clear success metric, honestly
 5. DECIDE      ship it · kill it · iterate
 6. REPEAT

The mindset that matters most: most experiments fail, and that's the point. The value is in cheap, fast, honest tests that kill bad ideas before they cost real money. A team running many small experiments learns faster than one betting big on a few.

Not fooling yourself: the hard part

Experimentation is worthless if you lie to yourself with it. The traps:

• Too small a sample. Tiny numbers are noise; a "20% lift" on 30 users means nothing. Respect statistical significance, or acknowledge you're reading tea leaves.
• Peeking and stopping early. Ending a test the moment it looks good manufactures false wins.
• HARKing (hypothesising after results): inventing the hypothesis to fit whatever the data did.
• Ignoring losers and cherry-picking winners. Confirmation bias with a dashboard.
• Local maxima. Endless small A/B tweaks polish a fundamentally weak product; sometimes you need a big swing, not another button colour.

The scientific method only works if you're willing to be proven wrong. An experiment you can't fail isn't an experiment, it's a justification.

Sequencing: retention → loop → scale

The correct order, and the one most founders get wrong:

 1. RETENTION FIRST   a curve that flattens — the product delivers durable value (Chapter 50)
 2. THEN A LOOP       a compounding acquisition/retention engine on top of it
 3. THEN SCALE        pour fuel (paid, effort) on a machine that already turns (Chapter 48)

Building a loop or buying scale before retention works is building on sand, you amplify a product that doesn't hold users.

📐 Best practice

Build loops, not just funnels, engines that turn after you stop pushing.

Only build a loop on top of working retention (Chapter 50).

Stack complementary loops where you can.

Use funnel thinking to diagnose leaks within the loop.

Run many small, cheap experiments, expect most to fail.

Form a real hypothesis before testing: change → effect → because.

Respect statistical significance, don't read noise as signal.

Decide the success metric and stopping rule before you start.

Be willing to be wrong, kill losers honestly.

Sequence: retention → loop → scale, never the reverse.

💀 Common mistakes

⭐ Building a growth loop on broken retention. Accelerating the leak; louder failure.

Only thinking in funnels, renting growth forever, never building an engine.

Guessing instead of testing, shipping opinions as if they were validated.

Reading noise as signal, a "win" on a tiny sample.

Peeking and stopping early, ending tests the moment they look good.

HARKing, inventing the hypothesis after seeing the result.

Cherry-picking winners, ignoring losers, confirmation bias with a dashboard.

Local-maxima optimisation, endless button tweaks on a weak product.

No success metric defined up front, so every result can be spun as good.

Scaling before the loop and retention work, fuel on an engine that doesn't turn.

The professional workflow

 1. CONFIRM RETENTION works first — a flattening curve (Chapters 49-50)

 2. IDENTIFY A CANDIDATE LOOP
    viral · content · paid · network — which fits the product?

 3. MAP THE LOOP — what output feeds the next input? Where does it leak?

 4. INSTRUMENT each step of the loop (Chapter 38)

 5. FORM A HYPOTHESIS — change → effect → because

 6. DEFINE success metric + stopping rule BEFORE testing

 7. RUN THE SMALLEST EXPERIMENT that gives real signal

 8. MEASURE honestly — significance, no peeking, no HARKing

 9. DECIDE — ship / kill / iterate

10. THEN SCALE the loops that compound (Chapter 48)

Tools, websites & costs

Experimentation tooling is free at early scale, and feature-flag/A-B systems are built into PostHog and similar. The discipline (honest hypotheses, significance) costs nothing but rigour.

Alternatives & trade-offs

Loops vs funnels. Not really rivals, loops are the growth engine (compounding, owned), funnels are the diagnostic lens (finding leaks). Build loops; use funnels to debug them. A founder who only thinks in funnels rents growth forever.

Many small tests vs few big bets. Many small tests learn fast and can trap you in local maxima (polishing a weak thing); big bets can escape local maxima and risk more per swing. Run mostly small tests for optimisation, and periodically step back to ask whether you need a big swing instead of another tweak.

Rigorous vs fast experimentation. Rigour (significance, proper samples) gives trustworthy answers slowly; speed gives directional answers fast and risks acting on noise. Early, with low traffic, you often can't reach significance, so be honest that you're making directional bets, not proven calls, and don't dress a hunch as a result.

Product-led vs marketing-led growth. Product-led loops (the product spreads or retains itself) are cheaper and more defensible but demand the right product; marketing-led growth is more universal and more rented. Prefer product-led loops where the product supports them; fall back to marketing-led where it doesn't.

Optimise vs pivot. Experimentation improves what exists; some products need a bigger change than any A/B test delivers. When months of tests barely move retention, the problem may be the product, not the button, know when to stop optimising and rethink (Part II).

Checklist

• Retention works before I build a loop or scale
• I think in loops (engines), not only funnels (diagnostics)
• My candidate loop is mapped and instrumented
• Every experiment starts with a real hypothesis
• I define the success metric and stopping rule up front
• I run the smallest test that gives real signal
• I respect significance and don't read noise as signal
• I don't peek, stop early, or HARK
• I kill losers honestly and expect most tests to fail
• I sequence retention → loop → scale

📓 Case Study: the loops that were designed but never turned

Project: SOLIS and its sibling content channel. The growth architecture was thought through; not one loop ever completed a single turn, and not one experiment was ever run, because there were no users and no traffic (Chapter 45).

⭐ Two loops were genuinely designed, the right instinct. The plan didn't rely only on a funnel:

• A content loop. The programmatic content engine (Chapters 46-47) was meant to turn the app's lessons into short-form video that would attract viewers, some of whom would become users, a classic content loop where the product's own material fuels acquisition. Designing acquisition as a loop built on existing content, rather than a series of one-off campaigns, is exactly the compounding mindset this chapter argues for.
• A retention loop inside the product. The streak-and-progress habit loop (Chapter 50) was a genuine retention engine, and retention loops feed growth loops (retained users refer, sustain content, and justify spend).

The architecture was sound on paper: a retention loop in the product, a content loop feeding it. That's a legitimate stacked-loop design.

⚠️ But the sequence was inverted, and no loop ever turned. This chapter's core rule is retention → loop → scale, and SOLIS had:

• Unproven retention, the retention loop was designed but never validated against a user (Chapter 50). So any growth loop stacked on top was, by this chapter's warning, building on sand.
• A content loop with no completed cycle, the engine was built, but nothing was published at scale, no viewer ever became a user, and the loop never closed once. A loop that never turns compounds nothing.
• No launched product for either loop to actually run through.

⚠️🚩 Zero experiments, the discipline was never exercised. Nothing in this chapter's experimentation method was practised on real data:

• No A/B tests, the onboarding funnel, the copy (Chapter 43), the pricing (Chapter 26) were all decided by reasoning and taste, never by experiment.
• No hypotheses tested against users, every growth and product decision is an untested belief.
• No significance to respect, because there was never a sample.

Interestingly, the product-design process did show a related discipline, the adversarial code reviews (Chapter 30) and the copy iterations (Chapter 43) were a form of "propose → challenge → revise." But that's internal critique, not experimentation against real user behaviour. Being willing to be proven wrong by a colleague is not the same as being proven wrong by users, and only the latter validates a growth loop.

🚩 Entirely hypothetical. The loops are well-designed hypotheses; the experimentation muscle was never used on real traffic. Whether the content loop actually acquires users, whether the retention loop actually retains, and which changes would improve either, all unknown, because nothing ever ran.

What generalises:

1. Design growth as loops, not campaigns, SOLIS got this right conceptually (content loop + retention loop, stacked).
2. But a loop that never turns compounds nothing. Design is not operation; the engine has to actually run, on a launched product, over real cycles.
3. Sequence retention → loop → scale. Loops stacked on unproven retention are built on sand.
4. Reasoning is not experimentation. Taste and internal critique decide fast; only tests against real users tell you if you're right. A growth loop is a claim about behaviour, and behaviour can only be measured, not argued.

Lessons

1. ⭐ Loops compound; funnels leak. Build engines that keep turning after you stop pushing; use funnels to diagnose their leaks.
2. A loop only works on top of retention. Sequence retention → loop → scale, never the reverse.
3. A designed loop that never turns compounds nothing. The engine has to actually run on a launched product.
4. Stack complementary loops, a retention loop feeding a content loop is a legitimate, powerful design.
5. Run many small, cheap experiments and expect most to fail, that's the point.
6. Form a real hypothesis and define success up front, an experiment you can't fail is a justification.
7. Don't fool yourself, respect significance, don't peek, don't HARK, don't cherry-pick.
8. Reasoning is not experimentation. Taste decides fast; only tests against real users validate a claim about behaviour.
9. Know when to stop optimising and pivot, endless tweaks can't fix a weak product.

Next: Chapter 52: Hiring, Contractors and Working with AI →