What is a CRO hypothesis
A CRO hypothesis is a statement you can check with an A/B test: it says what you will change on your website, who will see it, which metric you expect to improve and why you think it will happen, backed by some piece of data.
By Fernando Blanco Dosil · Updated November 2, 2026
Why "I think" is not a hypothesis
"I think a green button will sell more" is an opinion. It doesn't say why, for whom or how you will know whether you were right. If the test fails, you learn nothing; if it wins, you don't really know why either.
A good hypothesis makes you think before you build. And whatever happens, it leaves you with something learned: if it fails, you know that reason wasn't what was holding your customers back.
Bad hypothesis versus good hypothesis
The same idea, written two ways.
| Bad | Good | |
|---|---|---|
| Product page | "We need more photos" | Because 40% of returns are about size, we believe adding the size guide next to the size picker on mobile will increase orders, because it removes doubt before buying |
| Cart | "The cart looks ugly" | Because many people abandon the cart when they see the shipping cost, we believe showing "Free shipping over €40" on the product page will increase orders, because it removes the surprise at the end |
| Home | "Let's try a new banner" | Because visitors arriving from Instagram bounce off the home page, we believe showing them the products from the ad at the very top will increase clicks to product pages, because they find what they came for |
The data in the good hypotheses is illustrative; yours has to come from your analytics or your customers.
The parts of a good hypothesis
They don't all have to be perfect, but if one is missing, the test will be hard to read.
- The evidence: an analytics figure, a complaint that keeps coming up in customer service, a pattern in reviews.
- The change: concrete, something that can be built and seen.
- The audience: all visitors or a group, such as mobile, new visitors or people from a campaign.
- The primary metric: the one that decides the result, ideally close to the business.
- The reason: why you think people will behave differently.
How Bixlow does it
In Bixlow, each hypothesis is a brief filled in step by step (problem, change, audience and metric) and linked to the workflow that puts it live. You can write it yourself, let the AI agent interview you and draft it with you, or ask it to analyse your website and propose hypotheses with the evidence behind them. Hypotheses from that analysis arrive scored with ICE and ranked, and the agent builds the test once you approve it.