What a hypothesis does
A hypothesis is a prediction you can check. In marketing it connects three things: a change you plan to make, an outcome you expect, and a reason you expect it. Writing it down before the test forces you to be specific, keeps you from rewriting the goal once results arrive, and gives you something to learn from whether the variant wins or loses.
Many ideas start as hunches: “the page feels cluttered”, “people don’t trust the form”. A hunch is a good starting point. The work is turning it into a statement that a test can support or contradict.
A simple template
Use this structure:
Because we observed [evidence], we believe that changing [element] to [new version] for [audience] will cause [measurable outcome] to [change direction]. We will know this is true if [success rule].
Each part earns its place:
- Evidence. Where did the idea come from? Customer comments, support questions, analytics showing a drop-off, or a look at the page with fresh eyes. Evidence stops hypotheses being pure opinion.
- Change. One clearly described difference between control and variant.
- Audience. All visitors, new visitors, a particular device, or a particular email segment.
- Outcome. The single primary metric, such as form completions per visitor.
- Success rule. What counts as a meaningful improvement, and the sample size or run length you will wait for.
A weak and a stronger example
Weak: “A better headline will improve the page.”
It does not say what “better” means, what will improve, or why. Almost any result can be argued to fit.
Stronger: “Visitor comments suggest people are unsure what the service includes. We believe replacing the headline ‘Solutions for You’ with one that names the service and who it is for will increase the share of visitors who start the sign-up form. We will judge this on form starts per visitor after the planned sample is reached.”
This names the evidence, the change, the metric and the timing. It could clearly be wrong, which is exactly what makes it testable.
Make it falsifiable
If no result could ever prove your idea wrong, it is not a hypothesis. Avoid vague words such as “improve”, “engage” or “better” unless you define them. Prefer “increase the proportion of visitors who click the main button”. Decide in advance what result would make you drop the idea.
One change, one primary metric
Testing a single change makes the result readable. If you alter the headline, image and button together, a win cannot be traced to any one of them. Likewise pick one primary metric. You may monitor others as a safety check, for example unsubscribes in an email test, but name the judge before you start.
Choose a metric close to the goal
A click on a call to action is easy to count, and is a fine metric if the goal is clicks. If the real goal is purchases, a click may not reflect it. Pick the measure closest to the outcome that matters that you can still collect in enough volume. For low-traffic sites this involves trade-offs, which is why sample size is worth checking before you commit.
Decide the size of effect that matters
Ask what improvement would be worth acting on. If a change would need to lift sign-ups by a tenth to justify the effort, say so. This avoids celebrating tiny differences that may just be chance, and it feeds into your planning for how long to run the test. Understanding statistical significance helps here, since it tells you how to judge whether an observed gap is likely to be real.
Add a reason
“Because” matters. If the variant wins and you know why you expected it to, you learn something reusable about your audience. If it loses, the reasoning tells you which assumption to question. Without a reason, you only learn that one version beat another.
A checklist
Before launching, check that your hypothesis:
- is based on some evidence;
- describes one change;
- names the audience;
- states one primary metric and direction;
- includes a success rule and planned duration;
- could be proven wrong.
Where to put it
Store each hypothesis in your testing log along with the date it was written. That way you can compare predictions with results later and see how often your instincts are right. Most hypotheses turn out wrong or inconclusive, and that is normal. A clear prediction is simply the best way to make each test teach you something.