Growth hypothesis: template and examples for B2B teams

Most B2B teams have no shortage of ideas, only a shortage of ideas you can actually test. An idea like “we should improve our homepage” is not an experiment. It is a feeling. Writing a growth hypothesis means turning that feeling into a testable prediction: what do you expect to happen, for whom, and how will you know whether you were right. TL;DR: a strong hypothesis is falsifiable, tied to a single primary metric and based on an observation, not on a gut feel. In this article you get a fill-in template, examples and the pitfalls that make your experiments worthless.

A growth hypothesis is not an academic exercise. It is the beating heart of growth marketing as a system: without sharp hypotheses you test at random, and without structure you will never know afterwards why something did or did not work.

What makes a hypothesis a hypothesis?

The difference between an idea and a hypothesis lies in testability. A hypothesis predicts an outcome that can turn out to be wrong. That sounds obvious, but in practice teams constantly write down predictions that cannot fail.

Take “better copy increases conversion”. What is “better”? Which conversion? By how much? On which page, for which visitor? You can never disprove that sentence, so you learn nothing from it. A falsifiable hypothesis states up front the point at which your idea falls apart. If you expect a new headline to lift demo requests by a noticeable margin and it does not happen, you have learned something. That is a gain, even when the experiment “fails”.

Three properties separate a real hypothesis from a wish:

• Specific: it names a concrete change, a concrete audience and a concrete metric.
• Falsifiable: there is a measurable outcome that can contradict your prediction.
• Grounded: it starts from an observation, not from “I think that”.

The template: from observation to testable prediction

Use this fill-in structure for every hypothesis your team writes. It forces you to start at the observation and ends at a measurable outcome.

Because we saw that [observation from data or conversations], we expect that [change] for [audience or segment] leads to [expected change in primary metric M]. We know we are right if [measurable result within a time window]. We know we are wrong if [opposite outcome].

Every part does work:

The observation is your starting point and immediately the most important discipline. Never start at the solution. “We should add chat” is a solution looking for a problem. “We saw in conversations with sales that prospects often drop off because they do not get a quick answer to one specific question” is an observation. You can put several solutions up against that.

The change is what you are actually going to do. Keep it as small as possible. The bigger the change, the less you will know afterwards which part caused the effect.

The audience stops you measuring an average that describes nobody. A change can work for new visitors and backfire for returning leads. Name the segment.

The primary metric is the core. Pick one, and pick a metric that connects to revenue or pipeline: qualified leads, demo requests, pipeline value. Not bounce rate or time on page if they have nothing to do with your growth goal. Read why in steer on your north star metric.

The boundaries (“we know we are right/wrong if”) are what finally make the hypothesis falsifiable. Write them down before the experiment, not after. Otherwise you will talk your way around every outcome.

Three worked B2B examples

Example 1: lead quality on a demo form

Because we saw in CRM data that a large share of demo requests does not match our ideal customer profile, we expect that adding one qualifying question (company size) for visitors on the demo page leads to a higher share of qualified requests. We are right if the share of matching leads rises noticeably without the total number of qualified requests dropping. We are wrong if the number of qualified requests falls in absolute terms.

Note the tension in the outcome: more qualification can simply mean fewer requests. By naming both sides, you avoid celebrating a “win” that shrinks your pipeline.

Example 2: messaging on the homepage

Because we heard in customer conversations that prospects position us as “too broad”, we expect that a sector-specific headline for first-time visitors from one target sector leads to more click-throughs to the service page. We are right if the click-through rate for that segment is noticeably higher than the current version. We are wrong if the click-through rate stays flat or drops.

Example 3: nurturing over email

Because we saw that many leads stay silent for months after a first download, we expect that a series of three substantive emails for these cold leads leads to more reopened sales conversations. We are right if the number of recovered conversations rises. We are wrong if the series has no measurable effect on reopened conversations.

Notice that every example starts at an observation and ends at a metric that touches your acquisition cost or pipeline. That is no accident. A hypothesis that is not tied to a growth goal is fun to test but changes nothing about your results.

The pitfalls that make your experiments worthless

Starting at the solution. By far the most common mistake. You have an idea, and afterwards you invent an observation to justify it. Turn it around: first collect observations from data and conversations, then prioritise which ones are most worth tackling.

Multiple metrics as “primary” at once. If everything is your primary metric, nothing is. Pick one to steer on. Keep an eye on secondary metrics to spot side effects, but let your decision hang on that single one.

Vague margins. “Increases conversion” is not falsifiable. You do not have to promise an exact percentage, but you do have to agree up front on what counts as a meaningful difference for you, so you are not arguing about interpretation afterwards. At small volumes, statistical significance is a real problem too: a good-looking difference can be noise.

No time window. A hypothesis without an end date gets tested forever and never concluded. Agree up front on how long you measure.

Leaving out the failure condition. Without “we are wrong if”, you will almost always find a reason why you were right after all. That is not learning, that is convincing yourself.

How to put hypotheses into a rhythm

One good hypothesis is nice. A rhythm of hypotheses that follow each other is what makes growth predictable. Collect ideas continuously in a backlog, phrase them as hypotheses with the template above, and prioritise on expected impact, your confidence in the observation and the effort it takes to test. Then work in short cycles: one or a few experiments per period, each time with an outcome defined in advance.

The point is not to win every experiment. The point is to learn something every cycle that makes you sharper the next time. A team that consistently writes falsifiable hypotheses builds up, within a few months, a treasure trove of knowledge about what does and does not work with their audience. That is exactly the knowledge you cannot buy, and it is the difference between loose tactics and a working growth engine.

That discipline is hard to keep up on your own alongside your day-to-day work. As a growth marketing agency, we build this rhythm of hypotheses, experiments and learning into teams that want to grow structurally instead of running one-off actions. Want to talk through how to build this into your growth approach?

Book a no-obligation call and we will look at your first three hypotheses together.