Unfortunately, 6 of my 7 growth ideas have been unsuccessful. It used to upset me a lot. That is exactly what it’s all about.

When not operating a SaaS product with a formal growth experiment programme, one isn’t truly learning anything. You’re wasting time and money with your guesses.

What I’ve seen over and over is a team reads a great case study, makes a copy of that playbook, waits three months, and still can’t understand why nothing has happened. This strategy is not typically the problem. The issue is there’s no system beneath it, no way to discover what actually works out for your product, your customers, and your stage at this moment. 

What is a SaaS growth experiment framework? 

A SaaS growth experiment framework is a repeatable process for creating, prioritising, executing, and learning from structured tests throughout the growth funnel: acquisition, activation, retention, and revenue. It is a documented process based on data and continually improves over time, rather than a random guess game.

This article describes the framework that I use during practice. I’ll tell you why most growth efforts fail even without it, how to determine where to begin experimenting, what the five steps are for each experiment, a real experiment with real numbers, and, naturally, where it all ends up. 

Why Most SaaS Growth Efforts Fall Apart Without Structure

Most growth teams go straight to channels without comprehending the problem. They roll out a paid marketing campaign, create a referral programme, revamp the onboarding emails, and track results with traffic or signups. Not because any of this will result in revenue that lasts.

Between 2023 and 2026, median CAC payback at $5M–$50M ARR companies grew to 18 months. It is not a quirk of distribution. That indicates that most companies are buying customers for whom they do not have a proven and dependable system in place to convert and retain them.

This is exactly what I did in the beginning of my growth work. Our activation rate, or percentage of users who reached the core “aha moment,” was 34% while I was pushing hard on acquisition. We were filling a leaky bucket, and we were having a party about water entering the bucket.

The takeaway for your SaaS: If you can’t cite a tracked experiment that made one clear funnel metric better in the past 60 days, you are operating on assumptions, not facts. 

The Growth Experiment Framework: Start With a Diagnosis, Not an Idea

Here’s a mistake I see time and time again — teams begin experiments in their brainstorming process without knowing where the real problem is. This is like giving a prescription before any tests are performed.

The first step I take is to determine the largest percentage gap between two steps of the AARRR funnel—acquisition, activation, retention, revenue, and referral. Put the first experiment in that drop. Not the most exciting stage, not the one your CEO likes to sort out. The weakest one. 

If you’re interested in the metrics I use to track at each stage, check out my SaaS metrics guide; it gives you the numbers that will tell you where your funnel is leaking.

There’s one funnel stage I don’t do at any cost: One per quarter. As soon as you attempt to increase acquisition and retention, you can’t take credit for anything. You will not be sure what is the thing that has moved the needle or why it has moved the needle.

What I was wrong about initially: I was always assuming that the biggest issue is acquisition, as that is the most visible number on all the dashboards. Twice the real issue was activation, in different quarters. Users were registering, but then getting nothing. Getting rid of activation generated more retained revenue than anything I accomplished with acquisition efforts.

An implication for your SaaS: When you’re looking to map the funnel, find the largest drop, and go there first, regardless if it’s the drop you were expecting.

The first of the top five is Testing Without a Hypothesis. The first of the top five is Testing Without a Hypothesis. 

The Root Cause Most Teams Miss: Testing Without a Hypothesis

This is the part most teams skip, and it’s also where the framework actually earns its value.

Most “experiments” aren’t experiments. They’re changes made in hope. A button colour gets updated. A headline gets rewritten. A new onboarding step gets added. The change ships, someone checks the numbers a week later, and then… nothing conclusive. Because there was no hypothesis, no isolated variable, and no definition of success written down before anything went live.

Very few “experiments” are experiments. They are changes which are hoped for. The colour of a button is changed. A headline is not the same as it used to be. An additional onboarding step is added. After a week, someone sees the numbers and the change ships, but nothing is conclusive. Since there was no hypothesis, no isolated variable, and no definition of success that was written down before going live.

Before you touch a single pixel, three things should be in place for it to be a real experiment: 

A real experiment needs three things in place before you touch a single pixel:

• A specific hypothesis: “If we add a progress indicator to onboarding, more users will complete Step 3.”
• A single measurable outcome: “Step 3 completion rate, measured over 14 days.”
• A minimum detectable effect: the threshold you’ve decided in advance is worth acting on.

One notable point here is that, according to data compiled by both KeyBanc and OpenView SaaS Benchmarks, only 11–30% of SaaS companies achieve the Rule of 40 in 2025. That’s a broad spectrum, and the difference between the best and the worst is not primarily due to product quality. It’s an operational discipline. Those that have strict experiment processes wind up in the top quartile. If you’re interested in the unit economics of why this is relevant, check out my SaaS unit economics breakdown.

The implication for your SaaS: If you can’t formulate your hypothesis in one sentence that has a clear measurable outcome, the experiment is not ready to run! 

How to Run a SaaS Growth Experiment: Five Steps

This is exactly what I do: each step has an identified output – not a meeting – not a conversation – but a document and/or a decision. 

Step 1 – Formulate a hypothesis: Follow this structure: If we (change), then the (metric) will (improve by X) because (reason). Target Outcome: One sentence written. It takes 20 minutes. If it cannot be expressed in one sentence, then it has to be thought of more.

Step 2 – Specify the metric and the time period. One single measure, not composite scores or “overall engagement”. Pre-program the measurement window prior to the test. Usually takes 7-21 days to activate, with 30-60 days of retention. Output: One metric, one threshold, and one hard end date.

Step 3 – Assume a minimum viable sample size. To detect the signal, you must have enough data to identify the noise. When testing conversion rates, a good rule of thumb is to have at least 100 conversions for each variant before making any decisions. Having no traffic at the site won’t help in the experiment, regardless of the numbers. Output: A decision of go/no-go for running the test now or later.

Step 4 – Create the smallest version possible. Do not make the complete feature to test the concept. Before, I’ve tested onboarding changes with a sequence of emails with no lines of product code written. Output: a live rough, but it can be measured.

Step 5 – Document everything, including the failures. All experiments are recorded in a common log, including the hypothesis, the results, what the data revealed, and the next steps. Errors should be in the log as well. They eliminate directions and prevent the team from running the same dead-end test six months later. Input: one experiment card.

If you don’t know if your core product concept would benefit from this kind of experimentation or not, and you are still in the process of determining that, check out my post on validating your SaaS idea before coding; it’s the best place to start.

The implication for your SaaS: a set of rules in someone’s mind is not a set of rules. Write it down, or it’s lost when that person is gone. 

A Real Experiment from My Growth Work: What the Data Showed

As part of my growth efforts, I did an activation experiment in Q4 2025.

We had 34% of the new users who took 3 basic steps within their first 7 days. We hypothesised that a single contextual tooltip at first exposure to the core feature would increase 7-day activation by 7-8 percentage points. We thought a single contextual tooltip on first exposure to the core feature would increase activation by 7-8 percentage points over 7 days.

We’ve been developing the tooltip in just one afternoon. No engineering sprint, no roadmap entry, no committee sign-off. We deployed it to 180 new users over 14 days. It was deployed to 180 new users in 14 days.

The outcome – activation has increased from 34% to 41%. A 7-point lift – just under our 8-point target but still a suitable direction to act on. The more intriguing number: The 23 percentage points higher 60-day retention rate of users who activated in the first 7 days was more significant than that of those who did not.

1 afternoon of labour. Two weeks of training on the run. Three paid campaigns yielded more compound value than we did in the same quarter.

If you want to know how we chose what to test in the first place, I wrote about the full customer signal for that test: SaaS customer discovery frameworks.

Your takeaway for SaaS: Your best experiments are likely to be within the product itself, rather than in your marketing channels. 

Where This Framework Breaks Down

This is no different from every other framework: it doesn’t work in all circumstances. Let’s take a quick look at the three scenarios in which it is no longer helpful, and you need to be aware of them before you embark on the process. 

When your sample size is too small. 

For less than 200 MAUs per month, most of the activation and retention experiments won’t get results that are statistically reliable. As ProductLed’s research on experimentation velocity has proven, you need volume to find wins. What you get from forcing experiments on thin data is worse than what you do not get when you don’t run the experiments. Qualitative information from user interviews, session recordings, and customer calls is more valuable than any A/B test can be at this point. 

When there’s no agreed activation moment. 

No experiment will significantly improve it if your team cannot come up with a clear definition of what constitutes “activated” and what action, if any, indicates that a user has truly valued the product’s core value. Before the first test run, a shared definition is crucial, period. 

When you measure too early. 

Teams have called a 5-day experiment a failure because the numbers “looked flat”. Time is important in retention experiments. When a test is aborted due to impatience, it ends up giving false negative results, which are lethal to good ideas that might otherwise have been successful. 

Implication for your SaaS: This is too early for this framework if you are not at product-market fit. Attract the first 50 customers to pay you. The experiments will be more meaningful when you have a signal you want to optimise.

Advanced: Compounding Experiments Across Funnel Stages

After at least three runs with one stage, and once that run has stabilised, it’s time to add a second stage. This is the real world of compound growth, not by locating a huge lever but by making little tweaks here and there until each time they multiply throughout the funnel.

The sequence I would recommend is: activation, retention, then acquisition — because the leverage for doing activation in a SaaS business is the largest for the earliest stages, followed by retention, then acquisition. The vast majority of teams do this in the wrong order. Powerful improvements in acquisition are only reinforced by having downstream stages working.

I used that activation experiment in combination with a retention experiment the next quarter, which was a basic 14-day re-engagement protocol for users who activated but did not come back within 7 days; that’s my own growth work. These two experiments combined added up to a 60-day retention increase of 19% over six months. Neither was clever. Both were systematic.

Your implication for the SaaS: one optimised stage of your funnel is better, which means that your experiment is better. Optimised stage for a proper growth engine with 2 stages. 

What to Do With This Now

A SaaS growth experiment framework isn’t a tool you install or a template you download. It’s a discipline, designing before building, measuring before concluding, and documenting before moving on.

If you do nothing else after reading this, do these three things:

A SaaS growth experiment framework is not an application that you install or a template that you download. It’s a discipline, designing before building, measuring before concluding, and documenting before moving on.

After reading this, make sure to do these 3 things: 

1. Map out your funnel for the week. Determine the stage that has the greatest difference in values. That’s where your first experiment resides.
2. Write one hypothesis today. Before doing anything in the product or marketing stack, write one hypothesis today in the format: “If we [change], then [metric] will [improve by X] because [reason]”.”
3. Start an experiment diary. Begin an experiment diary; it could be as simple as a spreadsheet, and from this moment on, record all experiments, even the failed ones.

One honest admission: This framework will not tell you what to test. Once you have a good idea, it provides a secure system for checking well water. Creating that idea is another set of skills, and you need to be intimately connected with your users, not just your analytics dashboard. Truthfully, it’s the harder one.

If you have an activation rate under 40% and you’re not sure what is causing your drop-off, begin using the funnel diagnostic in my SaaS metrics guide. It will take about 30 minutes, and you will have a clear first experiment to run.