Measuring B2B leads with website experiments

Almost everything written about conversion optimisation assumes ecommerce: high volume, immediate conversion, revenue attached to the session. B2B has none of those properties, and applying ecommerce advice to it produces confident nonsense.

Volume. A B2B site might get 3,000 sessions a month and 40 form fills. At those numbers, detecting even a 30% change in form fills takes months.

Cycle length. The lead that arrives today closes in four months, if it closes. You cannot measure revenue impact within an experiment window, ever.

Quality is invisible at submission time. Two variants can produce identical lead counts and completely different pipelines. The variant that generated twice as many leads may have generated them from people who will never buy.

That third one is the dangerous one, because the metric that's easy to measure moves in the opposite direction to the metric that matters. Removing qualifying fields reliably increases submissions and reliably decreases quality.

Given you can't measure revenue and you barely have enough form fills, the workable approach is a chain of progressively harder metrics.

Primary: an engagement event that predicts intent. Pricing page views, documentation visits, case-study reads, calculator usage. These are frequent enough to reach significance and they correlate with buying intent much better than generic sessions.

Secondary: form submissions. Track it, but understand that at your volume it will rarely be conclusive on its own.

Guardrail: qualified-lead rate. This is the one that stops you optimising into a worse business. It needs your CRM to tell you which leads were qualified, and it needs weeks after the test to become readable — so it's a guardrail you check after deciding, and a reason to reverse a decision rather than a gate before making it.

That's an uncomfortable compromise and it's honest. In B2B you often ship on the leading indicator and verify against quality later.

The volume constraint has the same consequence as it does for small ecommerce brands, only more so: only large, structural changes are detectable.

What's worth testing in B2B:

• The offer on the form. "Book a demo" versus "Get a quote" versus "See pricing" attract completely different people in different quantities. These effects are large.
• Gating. Whether the pricing page requires a form at all. Among the biggest effects available on a B2B site, in both directions.
• Form length. Not one field — three versus eight. And always with the quality guardrail.
• Who the page speaks to. A page written for the evaluator versus one written for the budget holder. Large effects because they're different people with different questions.

What's not worth testing: button copy, field labels, the order of two paragraphs. You will never detect these, and pretending otherwise burns quarters.

B2B traffic is heterogeneous in a way that hurts: existing customers looking for docs, job applicants, competitors, students, and actual prospects, all in the same sessions number.

Segmenting to paid traffic, or to visitors who reached a product page, removes a lot of noise and can make an effect visible that was invisible in aggregate. In Bixlow those are audience conditions, so the excluded traffic never enters the experiment rather than being filtered out afterwards.

This is one of the few cases where narrowing the population can help despite the sample cost, because the excluded groups have near-zero conversion and were pure variance.

The endpoint of a serious B2B measurement setup is passing the experiment and variant identifiers through the form into your CRM, so that months later you can ask which variant produced the pipeline rather than which produced the form fills.

That's the only way to close the loop. It's worth doing once your programme is producing decisions regularly — before that, it's infrastructure for questions you're not yet asking.

• B2B breaks the standard model: low volume, long cycles, and quality invisible at submission.
• Use an intent-predicting engagement event as your primary metric; form fills rarely reach significance alone.
• Qualified-lead rate is a guardrail you check after shipping, not a gate before it. That's the honest compromise.
• Only large structural changes are detectable: the offer, gating, form length, who the page addresses.