CRO for B2B SaaS: The Signup Is Not the Sale

A conversion test may finish before the deals it generated close. Demo requests are an earlier measure, not a revenue result. Choose that measure with care and record which later outcomes it cannot yet establish.

Check eligible traffic on the pricing and demo pages before designing the test. The sample-size arithmetic is worked through on the conversion rate optimization page. We run it against your funnel so the detectable change and required sample are clear before work begins.

The lag is the design problem

Optimising a form for completions and reporting the result as a revenue lift skips the only step that mattered. So the metrics get agreed before traffic starts. Demo requests is the primary and it is read on the date the sample arrives. Qualified opportunity, and then closed revenue, are read later at dates fixed in advance. The sample is too small to call on its own, and the readout says so. When the two disagree, that disagreement is the finding.

What this rules out is a cadence that bills quarterly revenue impact every month. What it produces instead is a log. Each test, its proxy result on the day, and its pipeline result once the pipeline is old enough to read. A year of that log is the most valuable thing the programme owns, because it tells you which proxies at your company actually predict money.

One buying committee, two variants

Testing tools randomise per visitor or per browser. A software purchase involves an evaluator, a champion, a security reviewer and whoever signs. They arrive separately, often from the same company, sometimes from the same office network, and a visitor-level split scatters them across both arms. The one conversion event, whenever it lands, gets credited to whichever of them happened to fill in the form.

Where your CRM can identify the account, we randomise and analyse at account level. Where it cannot, we say so and treat the result as noisier than the tool’s confidence interval claims. Check how the chosen platform assigns accounts and handles correlated observations. Do not assume its default confidence interval accounts for your buying process.

Two funnels fighting over one pricing page

A self-serve visitor wants a price, a plan comparison and a button. A sales-led buyer wants evidence you serve companies their size, the compliance paperwork their security team will ask for, and a person to talk to. One page tries to serve both. A change that wins for one motion can lose for the other, the combined number sits flat, and nobody learns anything.

Separating them is usually the largest improvement available. It is a change we would ship outright, because it is too big to fit inside one hypothesis. Distinct paths from the ad and from the search result. A pricing page that carries both and leads with the one matching the segment. Then tests run inside each path, where a result means something.

The fields that lose the test and win the quarter

Removing a form field can change both completion and qualification. Measure them separately. A field for company size, use case or timing may help sales decide which enquiries to pursue.

So a form test carries two numbers from the day it is designed. Completions now, qualified opportunities later. Sometimes the shorter form wins both, because the qualification was happening inside the CRM anyway. Sometimes it doubles completions and fills the calendar with nothing. Both are real outcomes and only one of them would have been visible on the day the test ended.

The step that decides it happens after the login

For a trial-led product the conversion that pays is activation, and activation happens inside the application. The marketing site can change the promise, the signup flow and who arrives. It cannot change whether a new account reaches the moment the product becomes obvious.

Which means the useful version of this work crosses the boundary. The test spans a marketing page and the first-run experience in the product. That is a different repository, a different release train and usually a different team. We will scope it that way. Or we will tell you the marketing-side test will read flat, because the drop-off sits on the other side of the wall.

Correction, 23 September 2026: the earlier version treated a fixed sales lag and buying-committee size as industry facts. Tests now use the actual funnel.