What is hypothesis work?

Good hypotheses are not invented at a whiteboard and not copied from a "top 10 CRO tips" list. They come from evidence about your specific visitors. Five sources almost always give more than an ideas evening:

• Funnel data and drop-off. Where in the flow do you lose the most visitors? The step with the biggest drop-off is usually the most profitable place to form a hypothesis.
• Friction. Long forms, unclear steps, hesitation over price or shipping. Every point where the visitor has to make an effort is a candidate.
• Customer interviews. What customers themselves say made them hesitate often points straight at a hypothesis you would never have guessed from the numbers.
• Sales and support input. The questions that keep coming up in sales calls and support tickets are objections in disguise, and objections are raw material for hypotheses.
• Session recordings. Watching real visitors move around the page reveals friction that aggregated data hides: where the cursor hesitates, where someone scrolls past the thing meant to sell.

The common thread is that the hypothesis should come from your specific funnel, not from a generic list that worked for someone else. Nielsen Norman Group's work on user research↗ shows how much of what really blocks a conversion only becomes visible when you watch real users. The closer the evidence sits to your own visitors, the stronger the hypothesis, and the easier it is to score when it comes time to prioritise.

You will always have more hypotheses than you have traffic and time to test. Prioritisation decides the order, and it should be driven by potential business effect, not by what is most fun or easiest:

1. 01Gather the hypotheses in a backlog. Write them all in one place, each in the same hypothesis structure, so they can be compared directly instead of living in different people's heads.
2. 02Score with ICE or PIE. ICE stands for Impact, Confidence, Ease: how big the effect if it holds, how sure you are of the evidence, how easy it is to implement. PIE (Potential, Importance, Ease) is the same idea. Give each factor a score, say on a scale of 1 to 10, and the gut feeling becomes a number you can sort on.
3. 03Weigh in traffic and time. A hypothesis that needs a lot of traffic on a low-traffic page will take a while to give an answer worth trusting. Low feasibility and thin traffic push a hypothesis down the queue even when the effect would be large.
4. 04Choose the one with the biggest business effect if the hypothesis holds. Top of the queue is the hypothesis that, if it is right, means the most for the business and that you can actually test soon. Not the one that is easiest to build, and not the one the boss happens to like.

The scoring is not an exact science, it is a way to turn prioritisation into decision support, not status: a number to sort on rather than an opinion to argue about. CXL's work on prioritising experiments↗ goes deeper on the models. The goal is simple: the hypothesis that means the most for the business, and that can be tested soon, ends up on top.

Hypothesis work rarely gets stuck on a shortage of ideas. It gets stuck on five recurring mistakes, all of them avoidable:

• A design preference disguised as a hypothesis. "Make it green" is a taste, not an assumption. Without an expected, measurable consequence it is not a hypothesis, however confident the person proposing it sounds.
• No business KPI. A hypothesis tied to more clicks or longer time on page can win without a single krona moving. If it is not linked to something measured against the business, you are measuring the wrong thing.
• Too vague to be disproven. "The page should feel more trustworthy" cannot be tested, because no result can show it was wrong. A hypothesis has to be sharp enough to be able to fall.
• Template application. "Top 10 CRO tips" are hypotheses for someone else's funnel. They can inspire, but pasted straight into your backlog you end up testing someone else's site, not your own.
• Never logging the learning. A test that is not documented disappears. Did the hypothesis hold, did it fall, or was the answer incomplete? If that is not recorded you repeat the same ideas and the backlog never gets any wiser.

The common thread is discipline, not tools. Falsifiability↗, the idea that a claim must be able to be disproven to mean anything, is the single best filter against bad hypotheses: if it cannot be disproven, it is not ready to be tested. What separates hypothesis work that moves the needle from work that moves charts is that every hypothesis carries a business KPI and a learning that gets saved.

With us no optimisation starts with an idea, but with your data. The loop is easy to describe and hard to hold when time is short, and it is the discipline to hold it that we stand for:

• We start from your funnel and your numbers. The hypotheses grow out of drop-off, friction and what you hear from customers and sales, not out of a template. And you own the data: accounts, tracking and evidence are yours and come with you if you change supplier.
• Every hypothesis gets the structure if we do X, then Y happens, measured via Z, because Q, and is tied to a business KPI before it goes into the backlog. If it is not sharp enough to be disproven, it goes back.
• We score with ICE, weigh against traffic and time, and test what means the most for the business first. For how a winning hypothesis is then validated statistically, read on in the guide to A/B testing.
• Everything is documented in the monthly report: the hypotheses, the prioritisation, the result and what to test next. Decision support, not status.

An experienced hand gets done in an hour what an inexperienced one takes four to do, and in hypothesis work it shows in what never even reaches the queue: the ideas that cannot be disproven, do not carry a business KPI or are not worth the traffic. That is how a backlog becomes a way forward instead of a wish list. To see what your visitors actually get stuck on, see how Memorise works with conversion.