Growth & Experimentation: A Learning Path for Building Compounding Systems

Growth is often taught as a collection of tactics.

Referral program. A/B test. New channel. Viral mechanic. Pricing experiment.

This learning path teaches a different model: growth as a system of loops, metrics, hypotheses and cross-functional learning.

It is designed for operators who already understand funnels and want to move from local optimization toward compounding systems.

Prerequisites

You should understand:

• AARRR;
• funnel stages;
• activation and retention;
• basic CAC/LTV economics;
• conversion-rate measurement.

Recommended prior reading:

• Growth Loops: How to Build Self-Reinforcing Growth Systems;
• Experimentation-Led Go-to-Market.

Learning outcomes

At the end, you should be able to:

• distinguish funnels from loops;
• map a growth loop;
• identify loop speed and weak edges;
• separate optimization from strategic experimentation;
• write falsifiable hypotheses;
• prioritize by learning value;
• choose primary and guardrail metrics;
• interpret negative experiments;
• build an experiment library;
• connect growth work to business outcomes.

Module 1 — Move from funnel thinking to system thinking

Read Growth Loops.

Funnels are useful for diagnosing progression.

Loops answer a different question:

How does the output of one cycle create input for the next?

Examples:

Referral loop

customer → invite → new user → activation → new invite

Content loop

user/content creation → discovery → new user → more content

Paid reinvestment loop

spend → customer → margin → reinvested spend

Marketplace loop

more supply → better demand experience → more demand → more supply

Exercise

Choose one growth mechanism.

Define:

• input;
• action;
• output;
• reinvestment;
• conversion at each edge;
• cycle time.

If the output does not feed the next input, you have a process, not a loop.

Module 2 — Measure loop strength and speed

A loop can be weak for two reasons:

• low conversion;
• slow cycle time.

Suppose two referral loops produce the same number of successful invites per user.

One cycles every three days. One cycles every 30 days.

The first can compound faster.

Exercise

For the loop you mapped, write:

• Input → action — Conversion: · Time:
• Action → output — Conversion: · Time:
• Output → new input — Conversion: · Time:

Select the weakest edge.

That becomes a candidate growth problem.

Module 3 — Learn the difference between optimization and experimentation

CXL distinguishes growth experiments from optimization and A/B testing.

That distinction matters.

Optimization

Improve an existing system where the problem and objective are relatively clear.

Example:

• improve checkout completion.

Experimentation

Resolve uncertainty around a strategic assumption.

Example:

• test whether a different segment has materially stronger retention.

A/B testing

One method for controlled comparison.

It can support either optimization or experimentation.

Exercise

Take ten items from your growth backlog.

Classify them:

• optimization;
• strategic experiment;
• instrumentation/research;
• execution.

Many teams discover that almost everything they call an “experiment” is execution.

Module 4 — Write hypotheses that can fail

Read Experimentation-Led Go-to-Market.

A useful hypothesis includes:

• target segment;
• intervention;
• expected behavior;
• primary metric;
• guardrail;
• time horizon.

Bad:

Improve onboarding.

Better:

For new self-serve teams, showing the setup checklist immediately after signup will increase the share reaching the first collaborative project within 24 hours, without reducing 14-day retention.

Exercise

Write three hypotheses.

Then ask:

If this test is negative, what will we learn?

If the answer is “nothing,” rewrite the hypothesis.

Module 5 — Build the experimentation stack

Read Experimentation & CRO Stack: How to Turn User Behavior Into Measurable Growth.

The stack should connect:

1. behavioral evidence;
2. hypothesis backlog;
3. controlled delivery;
4. primary and guardrail metrics;
5. analysis;
6. rollout;
7. knowledge storage.

Do not buy experimentation software before the team has a hypothesis process.

Exercise

Audit your current system.

Where does each item live?

• insight;
• hypothesis;
• design;
• implementation;
• result;
• decision;
• learning.

Identify where information is lost.

Module 6 — Study iterative experimentation

Read TruckersReport: Six CRO Tests That Lifted Landing Page Conversion 79.3%.

The important lesson is not the historical 79.3% result.

It is that several plausible best practices lost.

The team learned from each round.

This is what a real experimentation program does:

`test → learning → next hypothesis`

rather than:

`test → winner screenshot → forget`

Exercise

Create an experiment record template containing:

• ID;
• hypothesis;
• evidence;
• audience;
• primary metric;
• guardrails;
• result;
• interpretation;
• next question.

Module 7 — Study local vs. global optimization

Read Smartlook: Why a 70%+ Signup Conversion Rate Was Not the Metric That Mattered.

Smartlook’s signup completion improved while activation deteriorated.

This illustrates why experiments need metric hierarchy.

Primary local metric

The immediate behavior.

Downstream metric

The next meaningful stage.

Guardrail

A measure that protects against damage.

For signup optimization:

• local: signup completion;
• downstream: activation;
• guardrail: support burden or retention.

Exercise

For your three hypotheses, add one downstream metric and one guardrail.

Module 8 — Know when the problem requires a strategic pivot

Read Pangea: When a Retention Problem Forced a Business Model Pivot.

Some problems cannot be solved through interface optimization.

If the retention curve is structurally weak, the hypothesis may need to move up a level:

• audience;
• use case;
• product;
• pricing;
• marketplace design;
• business model.

Growth operators need permission to escalate the problem.

Exercise

For the weakest metric in your system, write hypotheses at three levels:

Tactical

• change execution.

Product

• change experience or feature.

Strategic

• change segment, proposition or model.

Module 9 — Connect growth to North Star and input metrics

Reforge separates high-level outcome metrics from the input metrics teams can directly influence.

Amplitude similarly recommends starting from the customer value moment when defining a North Star Metric, then validating that it predicts retention and revenue.

Do not use one metric as a substitute for the full growth model.

Exercise

Define:

North Star / outcome What valuable behavior or economic outcome indicates healthy growth?

Inputs Which behaviors can teams influence this month?

Guardrails What must not deteriorate?

This creates a metric tree.

Final project — Build a growth experiment system

Your final deliverable includes:

Growth model

• one funnel;
• one loop;
• current constraint.

Metric tree

• outcome metric;
• three input metrics;
• two guardrails.

Backlog

At least ten hypotheses, scored by:

• impact;
• uncertainty/confidence;
• effort;
• learning value.

Experiment design

Write one full experiment:

• hypothesis;
• audience;
• treatment;
• primary metric;
• guardrails;
• duration;
• decision rule.

Knowledge loop

Define where completed experiments are stored and how they create the next hypothesis.

Suggested cadence

Day 1: loops
Day 2: loop measurement
Day 3: experimentation vs. optimization
Day 4: hypothesis design
Day 5: TruckersReport and Smartlook cases
Day 6: strategic escalation and Pangea
Day 7: metric tree and final experiment system

What to study next

Move to Measurement & Metrics if your main weakness is confidence in data or metric definitions.

Return to Acquisition Strategies if the growth constraint is new-customer volume.

Return to Funnels & Lifecycle if activation or retention is weak.

The goal of growth is not to run more tests.

It is to create a system that repeatedly converts uncertainty into evidence and evidence into stronger compounding behavior.

Build an experimentation portfolio, not only a backlog

A list of experiments can become biased toward whatever is easy to ship.

A stronger portfolio deliberately balances different kinds of uncertainty.

Conversion experiments

Improve an existing journey.

Examples:

• pricing page;
• checkout;
• onboarding.

Growth-model experiments

Test how one cohort can create the next.

Examples:

• referral;
• content loop;
• partner loop.

Market experiments

Test strategic assumptions.

Examples:

• new ICP;
• new use case;
• new geography;
• new channel.

Monetization experiments

Test:

• packaging;
• price presentation;
• trial structure;
• upgrade paths.

Measurement experiments

Improve confidence in the system itself.

Examples:

• validate activation definition;
• test incrementality;
• reconcile attribution.

Exercise

Take your top 15 backlog items.

Tag each by experiment type.

If more than two-thirds sit in one category, the team may be optimizing one layer while ignoring larger uncertainty elsewhere.

Experiment quality review

Before launch, ask:

• Is the hypothesis falsifiable?
• Is there enough traffic or sample opportunity?
• Is the primary metric close enough to the intended value?
• Are guardrails defined?
• Could another concurrent change contaminate the result?
• What decision changes if the result is positive?
• What decision changes if it is negative?
• Is the learning worth the implementation cost?

A test with no decision consequence is usually a weak test.

The strongest programs optimize for decision quality, not experiment count.