In marketing and website testing, a hypothesis is a specific, testable prediction: if this change is made for these people, this measurable result will move, for this reason. It is the written starting point of an A/B test, and it decides what you build, what you measure and how you read the outcome.
How a hypothesis works
A useful hypothesis has four parts: the evidence that prompted it, a single change, the expected effect on a named metric, and the reason you expect that effect. A simple template keeps them together:
Because [evidence], [change] for [audience] will [increase or decrease] [metric], because [reason].
A hypothetical example for a physiotherapy clinic in London: because session recordings show mobile visitors scrolling past the booking button to look for prices, showing the price of an initial assessment beside the button on mobile will increase completed bookings from paid search, because uncertainty about cost is holding people back.
Each part does a job. The evidence stops you testing whims. The single change means any difference can be traced to one cause. The metric, chosen in advance, stops you searching twenty metrics afterwards for one that happens to look good. The reason is what you learn whichever way the result goes: if showing the price makes no difference, cost was probably not the barrier, and you look elsewhere.
The hypothesis also feeds the planning. The size of effect you expect to detect sets the sample size you need, which in turn sets how long the test must run.
Why it matters
Many UK small and mid-sized websites do not get enough traffic to test lots of ideas, so every test costs weeks. A clear hypothesis makes sure those weeks are spent on something with a real chance of mattering, and that a losing test still teaches you something.
It also guards against the most common way testing goes wrong: changing several things, watching the numbers for a few days and declaring a winner when the line goes up. Writing the prediction and the success measure first, then judging the result against statistical significance, keeps decisions honest. The habit helps outside formal tests too. When you change ad copy, a landing page or an email subject line, one sentence written beforehand turns an edit into something you can learn from.
Common mistakes
- Writing a goal instead of a prediction. “Improve the homepage” cannot be proved wrong.
- Changing headline, image and form in one variant and treating it as one hypothesis, so nobody knows which change mattered.
- Choosing the success metric after seeing the results.
- Picking a metric too far from the change, such as total monthly revenue for a test of button wording.
- Leaving out the reason. Without it, a winning test cannot be applied elsewhere and a losing one teaches nothing.
- Stopping the test the moment it looks good, before the planned sample is reached.
How to act on it
Keep a testing log. A shared spreadsheet is enough, with one row per hypothesis: evidence, change, audience, metric, expected direction, reason, planned sample, dates, result and what you learnt. Over a year it becomes a record of what your customers respond to, which is worth more than any single result.
Gather evidence before writing anything: analytics reports showing where people drop out, recordings and heatmaps, customer service emails, sales call notes and reviews. Then rank the hypotheses by likely impact, how strong the evidence is and how much effort the change takes, and test the best first.
If traffic is too low for a reliable test, still write the hypothesis, make the change and compare a reasonable period before and after, cautiously and with seasonality in mind. Building landing pages around written hypotheses, and testing them where traffic allows, is how I run landing page design for paid campaigns.