A lightweight system for turning growth experiments into reusable decisions, not one-off readouts.
The goal is simple: every test should produce a result, a business impact readout, a core learning, a decision rule, and a clear next test.
This repo is the AI-readable source of truth for growth experiment learnings. It works alongside the Growth Learning Loop testing database in Google Sheets, Confluence readouts, and Slack summaries.
Structured testing database:
Growth Learning Loop Google Sheet
Use the Google Sheet for structured tracking and filtering.
Use this GitHub repo for clean markdown experiment records, reusable principles, and AI-readable context.
Most experiments die as isolated readouts.
The Growth Learning Loop turns each experiment into reusable company knowledge.
Each completed experiment should answer:
- What changed?
- What happened?
- Why does it matter?
- What did we learn?
- What decision rule should we use next time?
- What should we test next?
- Where else does this learning apply?
growth-learning-loop/
├── README.md
├── experiments/
│ └── homepage-builder-v1.md
├── templates/
│ ├── experiment-card-template.md
│ ├── slack-summary-template.md
│ └── executive-readout-template.md
├── assets/
│ └── figma-exec-readout-template/
├── reports/
│ └── homepage-builder-v1-exec-readout.md
└── principles/
└── growth-decision-rules.md
Every completed experiment should include:
- Test name
- Surface
- Owner
- Date range
- Status
- Hypothesis
- Strategic context
- Control
- Variant
- Primary metric
- Secondary metrics
- Result
- Key lifts
- Business impact
- Core learning
- Decision rule
- How to read the result
- Next test
- Where else this applies
- Links to dashboard, screenshots, Confluence, and the Google Sheet row
A good experiment learning is not just:
The variant won.
A good experiment learning sounds like:
Builder-first did not mean builder-only. Clearer positioning helped every audience convert, including enterprise.
The best learnings become decision rules.
Example:
Before making a page sound more “enterprise,” ask whether we mean more credible or just more jargon-heavy.
Every completed experiment should create four outputs:
-
Database row
A structured record in the Growth Learning Loop Google Sheet. -
Experiment markdown file
A clean AI-readable record in theexperiments/folder. -
Executive readout
A concise Confluence-ready summary with the result, business impact, learning, and next step. -
Slack learning post
A short post that shares the learning and asks the team where else it should apply.
What happened?
Use clear numbers. Show control, variant, and change when possible.
Example:
| Metric | Control | Variant | Change |
|---|---|---|---|
| AI setup starts | 4.9% | 10.5% | +114% |
| Enterprise inquiry CVR | 11.9% | 16.7% | +40.3% |
| Self-serve signup CVR | 6.6% | 7.7% | +17% |
Why does it matter commercially?
Separate observed results from modeled projections.
Example:
Using a conservative 25% sustained inquiry lift, this homepage change models to ~+56 MQLs, ~7 Stage 2 opportunities, and ~$950K in projected pipeline over the next 6 months.
What did we learn that should influence future decisions?
Example:
Builder-first did not mean builder-only. Clearer positioning helped every audience convert, including enterprise.
What should we do differently next time because of this test?
Example:
Before making a page sound more “enterprise,” ask whether we mean more credible or just more jargon-heavy.
What should we test next?
Example:
Hero language + CTA architecture.
What other surfaces should use this learning?
Examples:
- Homepage
- Pricing page
- Paid search landing pages
- AI setup pages
- Startup program pages
- Enterprise demo pages
- Lifecycle upgrade pages
When using ChatGPT, Claude, or another AI tool, provide the relevant experiment markdown file and ask it to generate:
- Executive summary
- Slack post
- Confluence readout
- Portfolio case study
- Next-test recommendations
- Decision rules
- Applications to other pages or campaigns
Suggested prompt:
Use this Growth Learning Loop experiment record to create:
1. a punchy executive summary,
2. a Slack-ready growth learning post,
3. a Confluence readout,
4. recommended next tests,
5. a reusable decision rule.
Keep the tone clear, strategic, plainspoken, and data-driven.
Do not overclaim.
Separate observed results from modeled projections.
Make the learning useful beyond this one test.
End with where else this learning should apply.
When analyzing an experiment, follow these rules:
- Lead with the business-relevant result.
- Do not treat every lift as equally important.
- Separate observed data from modeled impact.
- Do not overclaim causality when the result only supports a directional interpretation.
- Always identify the strategic learning, not just the winning variant.
- Always turn the learning into a decision rule.
- Always recommend the next test.
- Always ask where else the learning applies.
- Prefer clear, human language over internal jargon.
- Make the output useful for executives, operators, and future AI agents.
Use this style for all readouts and summaries:
- Plainspoken
- Punchy
- Specific
- Data-driven
- Commercially aware
- No corporate jargon
- No vague words like “interesting,” “directional,” or “promising” unless needed for accuracy
- No inflated claims
- No burying the business impact
- No “variant performed better” without explaining why it matters
Good:
Builder-first did not mean builder-only. Clearer positioning lifted self-serve, AI setup, pricing intent, and enterprise conversion.
Weak:
The new homepage variant performed well across a number of key metrics and may indicate a positive direction for future testing.
Use this format when sharing learnings in Slack:
New growth learning: [Experiment Name]
Result:
- [Metric/result]
- [Metric/result]
- [Metric/result]
Learning:
[The core learning in one or two sentences.]
Business impact:
[Pipeline, signup, activation, revenue, or efficiency impact. Separate modeled projections from observed results.]
Decision rule:
[What we should do differently next time.]
Next test:
[Recommended follow-up test.]
Question:
[One sharp question that helps the team apply this learning elsewhere.]
Example:
New growth learning: Homepage Builder V1
Result:
- +114% AI setup starts
- +40% enterprise inquiry CVR
- +17% self-serve signup CVR
- +35% pricing traffic
Learning:
Builder-first did not mean builder-only. Clearer positioning helped every audience convert, including enterprise.
Business impact:
Using a conservative 25% sustained inquiry lift, this models to ~$950K in pipeline over the next 6 months.
Decision rule:
Before making a page sound more “enterprise,” ask whether we mean more credible or just more jargon-heavy.
Next test:
Hero language + CTA architecture.
Question:
Where else are we hiding value behind enterprise language?
Use this structure for Confluence or leadership updates:
-
Headline
One punchy line that captures the strategic learning. -
Summary
Two to three sentences explaining what changed, what happened, and why it matters. -
Key results
A small table with the highest-signal metrics. -
Business impact
The commercial impact or modeled projection. -
What we learned
Three punchy takeaways. -
How we are reading it
Caveats, nuance, or quality checks. -
Recommended next step
The next test or decision. -
Decision rule
The reusable principle from the test.
For designed executive one-pagers, use the template package in:
assets/figma-exec-readout-template/
Source experiment: Homepage Builder V1
Learning:
Clearer, less jargon-heavy positioning lifted conversion across self-serve, AI setup, pricing intent, and enterprise inquiry CVR.
Decision rule:
Before making a page sound more “enterprise,” ask whether we mean more credible or just more jargon-heavy.
Why it matters:
There was concern that a builder-first homepage would hurt enterprise demand. The test showed the opposite. When the page made the value clearer and easier to understand, enterprise conversion improved too.
Applies to:
- Homepage
- Pricing page
- Paid search landing pages
- AI setup pages
- Startup program pages
- Enterprise demo pages
- Lifecycle upgrade pages
Watchouts:
Validate downstream quality before scaling the learning too broadly. For sales paths, check MQL rate, Stage 2 rate, pipeline created, and segment mix. For self-serve, check activation and retention signals.
Do not treat experiments as status updates.
Treat them as a learning system.
Each test should make the next decision easier.