How to Write a Good Test Hypothesis

A good hypothesis is the difference between a test that generates learning and one that just burns through traffic. Without a hypothesis, you test changes at random and, whether you win or lose, you don't know why it happened or what to do next. With a hypothesis, every experiment answers a concrete question and adds to your knowledge of your audience.

Writing hypotheses doesn't take special talent, it takes method. You need an observed problem, a proposed solution, and a measurable prediction. In this article you'll see how to structure them, where to find good ideas, and which signals tell you a hypothesis is poorly framed before you waste weeks on a useless test.

The structure of a solid hypothesis#

A useful hypothesis follows a simple pattern: if we change X, then Y will happen, because Z. The "if" part describes the concrete modification, the "then" part predicts the effect on a measurable metric, and the "because" part explains the reasoning based on data or observed user behavior. That last piece is the one many people skip, and it's the most valuable.

For example: if we cut the signup form from six fields to three, then the signup rate will increase, because heatmap data shows users drop off when they reach the fourth field. A hypothesis like this tells you what to measure, what to expect, and what you'll learn whether you're right or wrong.

Where good hypotheses come from#

The best hypotheses aren't invented in a meeting, they come from evidence. The more tightly your hypothesis is connected to real data, the higher the odds the test will move the needle. Combine several sources to spot where the friction points are.

• Web analytics: pages with high drop-off, funnels with steep declines, or steps with low conversion.
• Heatmaps and session recordings: where users click and where they get stuck.
• Surveys and interviews: what doubts or frustrations your visitors express.
• Customer support: repeated questions that reveal unclear information on the site.

Signs your hypothesis is failing#

If you can't define which concrete metric you'll measure, your hypothesis is too vague. If the change you're proposing is so small the user won't even notice it, the impact will be imperceptible. And if you can't explain why you think it will work, you're probably just guessing. Check these three points before launching and discard any hypothesis that doesn't pass them.

Key takeaways#

• Structure every hypothesis as: if we change X, then Y, because Z.
• Ground your hypotheses in real data from analytics, heatmaps, or customer support.
• Discard hypotheses without a clear metric, without expected impact, or without reasoning.