Growth experiments: the complete guide for startups
Everything you need to know about growth experiments. Learn the framework, prioritization methods, and proven playbooks that startups like Dropbox, Airbnb, and Slack use to grow systematically.
If you are a solo founder or indie hacker, you have probably felt the pain of guessing what to work on next. You ship a feature, tweak some copy, run an ad, and hope something sticks. But hope is not a strategy. Growth experiments give you a structured way to find what actually works, without burning through your runway on hunches.
This guide covers everything you need to know about growth experiments: what they are, how to run them, how to decide which ones to run first, and how to measure whether they worked. It is written specifically for people who do not have a dedicated growth team or a data science department. Just you, your product, and a willingness to learn fast.
We will reference real examples from companies like Dropbox, Airbnb, Slack, Duolingo, and others, not because you need to be like them, but because their experiments are well-documented and the principles translate to any stage.
What are growth experiments?
A growth experiment is a structured test designed to improve a specific business metric. You start with a hypothesis like “if we simplify the signup form from five fields to two, more visitors will create accounts.” Then you design the test, run it, measure the result, and learn from it. That learning part is the most important bit. Even failed experiments teach you something about your users.
Growth experiments are not the same as traditional A/B tests, though there is overlap. An A/B test typically compares two versions of one element, like a headline or button color. Growth experiments are broader. They can involve launching an entirely new onboarding flow, testing a referral program, trying a new acquisition channel, or rethinking your pricing model. The key difference is the scope and the intent. A/B tests optimize what exists. Growth experiments explore what could exist.
Startups use growth experiments because they cannot afford to build the wrong thing. When Dropbox was trying to grow in 2008, they did not just run ads. They ran a referral experiment that gave both the referrer and the friend extra storage. That single experiment drove 60% of their signups and helped them grow from 100,000 to 4 million users in 15 months. They did not guess. They tested.
Airbnb ran experiments on their listing photos, discovering that professional photography dramatically increased bookings. Slack experimented with activation metrics, finding that teams who sent 2,000 messages almost always converted to paid customers. None of these insights came from brainstorming sessions. They came from structured experimentation.
The beauty of growth experiments is that they work at any scale. You do not need millions of users. Even with a hundred users, you can run meaningful experiments. The approach changes (qualitative over quantitative at small scale), but the principle stays the same: form a hypothesis, test it, learn from it, repeat. If you want a deeper comparison, read our guide on growth experiments vs A/B tests.
The growth experiment framework
Every growth experiment follows the same five-step cycle: hypothesis, design, execute, measure, learn. This framework is not just theory. It is the actual process used by growth teams at companies of all sizes, and it is how solo founders can punch above their weight. Let us walk through each step.
Step 1: Form a hypothesis
A hypothesis is a specific, testable prediction about what will happen if you change something. It forces you to think before you build. A bad hypothesis is “let us try a new landing page.” A good hypothesis is “if we add social proof (customer logos and testimonials) above the fold, our visitor-to-signup conversion rate will increase by 15% because visitors currently lack trust signals.”
The structure matters. Your hypothesis should include what you are changing, what metric you expect to move, by how much, and why you believe it will work. The “why” is crucial because it connects the change to user behavior. If the experiment fails, the “why” helps you understand what you got wrong about the user. Without it, you just know something did not work, which is not very useful.
Where do hypotheses come from? The best ones come from user data and conversations. Look at where users drop off in your funnel. Read support tickets. Watch session recordings. Talk to users who churned. The pattern you notice becomes your hypothesis. If you are just getting started, our guide on running your first growth experiment walks you through the process step by step.
Step 2: Design the experiment
Design means deciding how you will test your hypothesis. This includes what you will change, who will see the change, how long you will run the test, and what counts as success. Keep the scope small. The biggest mistake founders make is designing experiments that take weeks to build. If your experiment requires a full rewrite of your onboarding, scope it down. Test the riskiest assumption first.
For example, if your hypothesis is that a guided onboarding will improve activation, do not build a full interactive walkthrough. Instead, test the concept by manually sending new users a personal email with three steps to get started. If that manual version moves the metric, then invest in building the automated version. This is the playbook approach: start with the smallest version that tests the core assumption.
Define your primary metric (the one thing you are trying to move) and one or two guardrail metrics (things that should not get worse). If you are testing a more aggressive upgrade prompt, your primary metric might be free-to-paid conversion rate, and your guardrail might be user satisfaction or churn rate. You do not want to increase conversions by annoying everyone else.
Step 3: Execute
This is the build-and-ship phase. For solo founders, speed matters here. The faster you can get the experiment live, the faster you learn. Do not over-engineer the implementation. Use feature flags, manual processes, or quick-and-dirty code. The experiment is temporary. You can clean it up later if it wins.
One tip that saves a lot of grief: document what you are changing before you change it. Screenshot the current state, note the current metrics, and write down the exact change you are making. Future you will thank present you when it is time to analyze results and you cannot remember what the old version looked like.
Step 4: Measure
Let the experiment run long enough to get meaningful data. How long depends on your traffic. If you get 100 signups per day, a week might be enough. If you get 10 per week, you will need a month or more. The key is not to peek at results early and make decisions based on incomplete data. Set a duration in advance and stick to it.
For small-scale experiments (fewer than a few hundred users), do not worry about statistical significance calculators. Use directional data combined with qualitative signals. Did the metric move in the right direction? Did users mention the change positively? Did support tickets change? At early stages, a clear directional signal plus user feedback is more valuable than a p-value. For more on this, read our guide to measuring experiment results without a data team.
Step 5: Learn
This is the most underrated step. After you measure the results, write down what you learned. Not just “it worked” or “it failed,” but why. What does this tell you about your users? What assumptions were validated or invalidated? What should you test next based on this?
The compound effect of growth experiments comes from accumulated learning. Each experiment teaches you something about your users. Over time, your hypotheses get sharper because you understand your users better. The companies that grow fastest are not the ones that run the most experiments. They are the ones that learn the most from each experiment. If an experiment fails, read our guide on what to do when an experiment fails — it is often more valuable than a win.
How to prioritize experiments (ICE scoring)
You will always have more experiment ideas than time to run them. This is a good problem, but it still needs solving. The ICE scoring framework is the most practical prioritization method for small teams because it is fast and it works.
ICE stands for Impact, Confidence, and Ease. For each experiment idea, score each factor from 1 to 10. Impact: if this experiment succeeds, how much will it move the metric? A change that could double conversions gets a 10. A change that might improve things by 2% gets a 3. Confidence: how sure are you this will work? An experiment based on strong user feedback and industry precedent gets an 8 or 9. A pure gut feeling gets a 2. Ease: how quickly can you implement and run this? Something you can ship in a day gets a 9. Something that takes a month gets a 2.
Average the three scores and rank your experiments from highest to lowest. Run the top-scored experiment first. The magic of ICE is that it balances ambition with pragmatism. High-impact but hard experiments get balanced by their ease score. Easy but low-impact experiments get balanced by their impact score. You end up focusing on experiments that are meaningful and achievable.
A common mistake is inflating confidence scores. Be honest with yourself. If you have no data supporting an idea, give it a low confidence score. That is fine. Low confidence just means you need a smaller, faster test to validate the assumption before investing more. For a detailed walkthrough with examples, check out our full ICE scoring guide.
There are other frameworks too (RICE, PIE, BRASS), but ICE is the best starting point because it requires the least overhead. You can score experiments in a spreadsheet or even on paper. The point is not the framework itself. The point is having a repeatable way to make decisions instead of going with whatever feels exciting today.
The AARRR framework
The AARRR framework (also called pirate metrics because, well, say it out loud) organizes your growth experiments into five stages of the user journey. It was created by Dave McClure and has become the default mental model for startup growth. Understanding it helps you know where to focus your experiments.
Acquisition
How do people find you? This covers SEO, content marketing, paid ads, viral loops, partnerships, and any other channel that brings new visitors. The key metric is usually visitor-to-signup conversion rate. If nobody knows you exist, nothing else matters. But once acquisition works, stop pouring more in and fix the rest of the funnel first.
Activation
Do new users get value quickly? Activation is the moment a user first experiences the core value of your product. For Slack, it was sending 2,000 messages. For Dropbox, it was saving a file. Activation experiments focus on reducing friction between signup and that first value moment. This is often the highest-leverage area for early-stage products.
Retention
Do users come back? Retention is the foundation of sustainable growth. If users leave as fast as they arrive, improving acquisition just fills a leaky bucket faster. Retention experiments focus on habit formation, engagement loops, re-engagement, and ongoing value delivery. Duolingo famously used streak mechanics to 3x their retention.
Revenue (monetization)
How do you make money? Monetization experiments cover pricing, free-to-paid conversion, upselling, and reducing payment friction. The timing matters: optimize monetization after you have good retention, not before. Slack found that usage-based upgrade prompts (showing the prompt when teams hit the message limit) dramatically outperformed random prompts.
Referral
Do users tell others? Referral experiments create loops where existing users bring new users. The viral coefficient (K-factor) measures this: if each user brings more than one new user, you get exponential growth. Referral only works when users genuinely love the product, so fix activation and retention first.
The most common mistake with AARRR is working on the stages in order. You do not need to perfect acquisition before touching activation. Instead, find your biggest leak. If you are getting signups but nobody activates, focus on activation. If people activate but churn after a week, focus on retention. Fix the worst leak first, regardless of the stage order. Our AARRR pirate metrics guide goes deeper into how to apply this framework to your specific situation.
Common growth problems
Most growth problems are not unique. Thousands of startups face the same challenges. The good news is that means there are proven solutions. Recognizing which problem you have is the first step toward finding the right experiment to run.
Below are six of the most common growth problems we see across startups. Each one has specific symptoms you can diagnose, root causes to investigate, and proven experiments that address it. The most effective approach is to identify which problem is costing you the most users or revenue right now, and focus your experiments there.
Users sign up and disappear
Users try once and never come back
Paying for every user when product should spread itself
Nobody uses your referral program
Users ignore upgrade prompts
Users drift away and forget you exist
These are just some of the growth problems you might face. Each problem page includes symptoms to help you confirm the diagnosis, root causes to investigate, quick wins you can try today, and linked playbooks with step-by-step experiment instructions. Browse all growth problems to find what matches your situation.
Proven growth playbooks
A playbook is a reusable experiment template based on what actually worked for a real company. Instead of starting every experiment from scratch, you can adapt a proven approach to your context. Think of it as a recipe. The ingredients might change, but the method has been tested.
For example, the Dropbox two-sided referral playbook documents exactly how Dropbox structured their referral incentives, where they placed the prompts, and how they measured success. If you want to build a referral program, you do not have to reinvent the wheel. You take their playbook, adapt it to your product and audience, and run the experiment. The core insight (two-sided rewards outperform one-sided rewards) is already validated. Your job is to test whether it works in your specific context.
Playbooks exist across every stage of the AARRR funnel. Acquisition playbooks from companies like Calendly (viral loops through calendar links), Notion (template marketplaces for SEO), and HubSpot (free tools that generate leads). Activation playbooks from Slack (identifying the right activation metric), Superhuman (concierge onboarding), and Pinterest (interest-based onboarding). Retention playbooks from Duolingo (streak mechanics), Spotify (annual recaps), and Amazon Prime (bundled value). Each one is a starting point, not a rigid prescription.
The best way to use playbooks is to match them to your current biggest problem. If users sign up but disappear, look at activation playbooks. If retention is solid but growth is slow, look at acquisition or referral playbooks. Do not pick playbooks based on what sounds interesting. Pick them based on where your funnel leaks the most. Browse all playbooks to find one that matches your situation.
How to measure results
Measuring experiment results is where a lot of founders get stuck. You do not need a data team or a statistics degree. You need to track the right metric, wait long enough, and be honest about what the data says.
Start with your primary metric. This is the one number your experiment is trying to move. Be specific: not “engagement” but “Day 7 retention rate.” Not “conversions” but “visitor-to-signup rate on the pricing page.” Vague metrics lead to vague conclusions. If you cannot put a number on it, you cannot measure it.
Next, add one or two guardrail metrics. These are metrics that should not get worse as a result of your experiment. If you are testing a more aggressive paywall, your primary metric is free-to-paid conversion rate, and your guardrail might be monthly active users. If the conversion rate goes up but active users tank, that is not a win.
For statistical significance, the rule of thumb is this: you need at least 100 users in each variation to have any confidence in the result, and ideally 300 or more. If you have less than that, rely on qualitative data. Talk to users. Watch them use the product. Read their messages. A 50% improvement noticed in 30 users is directionally meaningful even if it is not statistically significant.
Common pitfalls: stopping experiments too early (seeing an initial spike and declaring victory), not accounting for novelty effects (users engage with anything new, but the effect fades), and cherry-picking metrics after the fact. Decide your success criteria before you start. Write it down. Stick to it. For a full breakdown with practical examples, read our guide to measuring experiment results.
Key Metrics by Funnel Stage
If you want to learn the definitions and formulas for all these metrics, our growth metrics glossary has everything you need. And if you want to go deeper on picking the right metric for your stage, check out the north star metric guide.
Tools for growth experiments
You do not need expensive tools to run growth experiments. Many founders start with nothing more than a spreadsheet to track hypotheses, a basic analytics tool like Plausible or Google Analytics, and their product's own database. At early stage, your biggest bottleneck is not tooling. It is discipline: actually running experiments consistently instead of just building features.
That said, the right tools can remove friction and help you experiment faster. Here is what the typical stack looks like for a solo founder or small team:
- Experiment tracking: Something to log hypotheses, results, and learnings. This could be a spreadsheet, Notion, or a dedicated tool like Golden Gecko.
- Analytics: Google Analytics, Plausible, Mixpanel, or PostHog for tracking user behavior and funnel metrics.
- Feature flags: LaunchDarkly, Flagsmith, or even a simple database toggle to control who sees what.
- Session recording: Hotjar, FullStory, or PostHog to watch how users interact with changes.
- User feedback: Intercom, Canny, or even a simple email to capture qualitative data.
Where Golden Gecko fits in: most of these tools help you execute experiments, but they do not help you decide which experiments to run or what results mean. That is the gap Golden Gecko fills. It gives you a library of proven playbooks matched to your goals, AI that helps you interpret results, and a structured way to capture learnings so you build institutional knowledge over time. Think of it as the layer that sits on top of your analytics and turns data into decisions.
If you are evaluating tools, our comparison pages break down how different tools stack up for growth experimentation specifically.
Getting started
You now have the full picture: what growth experiments are, the framework for running them, how to prioritize, where to focus, and how to measure results. Here is a three-step action plan to start this week.
Find your biggest leak
Look at your funnel from acquisition to retention. Where are you losing the most users? That is where your first experiment should focus. Use the growth problems page to diagnose your specific situation. If you are not sure where to start, activation is usually the highest-leverage area for early-stage products.
Pick a playbook and adapt it
Browse the playbook library for an experiment that matches your problem. Read how it was originally implemented, then adapt it to your product and audience. Write a clear hypothesis with a specific metric and target. Use ICE scoring if you have multiple ideas competing for your attention.
Run it and write down what you learn
Ship the smallest possible version of the experiment. Let it run long enough to get meaningful data (at least a week for most experiments). Then write down what happened, what you learned about your users, and what you will test next. The learning is the whole point. Repeat this cycle and you will be running a genuine growth engine within a month.
The hardest part is starting. Your first experiment does not need to be perfect. It just needs to be intentional. Move from “I hope this works” to “I am testing whether this works.” That shift in mindset is the real value of growth experimentation.
If you are wondering how many experiments to aim for, our guide on how many experiments to run per month has practical guidance for different team sizes and traffic levels. And if you want to build this into a sustainable habit, read about building a growth experiment culture as a solo founder.
Frequently Asked Questions
What is a growth experiment?
A growth experiment is a structured test designed to improve a specific business metric. Unlike random A/B tests, growth experiments start with a hypothesis about user behavior, define a clear success metric, run for a set duration, and produce learnings regardless of outcome. They are the core tool of systematic growth, used by startups like Dropbox, Slack, and Airbnb to discover what actually moves the needle.
How many growth experiments should a startup run per month?
Most solo founders and small teams should aim for 2-4 experiments per month. Quality matters more than quantity. Each experiment needs enough time to produce meaningful results, typically 1-4 weeks depending on your traffic volume. Running too many experiments at once makes it impossible to attribute results to specific changes.
What is the difference between a growth experiment and an A/B test?
An A/B test compares two versions of something (like a button color) to see which performs better. A growth experiment is broader: it starts with a hypothesis about user behavior, can involve entirely new features or strategies, and focuses on learning. Every A/B test is technically an experiment, but not every growth experiment is an A/B test. Growth experiments can include new onboarding flows, pricing changes, referral programs, or entirely new acquisition channels.
How do I prioritize which growth experiments to run first?
The most popular framework is ICE scoring: rate each experiment on Impact (how much will this move the metric if it works), Confidence (how sure are you it will work, based on data or precedent), and Ease (how quickly can you implement it). Score each factor 1-10 and average them. Start with experiments that score highest. Also consider where your biggest funnel leak is - fix the leakiest part first.
What metrics should I track for growth experiments?
Track one primary metric per experiment (the thing you are trying to improve) and 1-2 guardrail metrics (things that should not get worse). Common primary metrics include conversion rate, activation rate, retention rate, revenue per user, and referral rate. Always define your success metric before running the experiment, not after. Use the AARRR framework (Acquisition, Activation, Retention, Revenue, Referral) to organize your metrics across the funnel.
Stop Guessing. Start Experimenting.
Golden Gecko gives you proven playbooks matched to your goals, step-by-step guidance, and AI that tells you what results mean. Always know what to test next.