A B2B website can have enough visitors to reveal confusing copy, broken forms and missing buying information while producing too few enquiries for a useful split test. The work can still move forward. Each decision needs a method suited to the evidence available.
Conversion rate optimisation, or CRO, includes investigating why intended buyers struggle to take a useful next step. A/B testing is one way to evaluate a proposed change. Reviewing enquiry quality, watching someone use the site and checking a form from submission through to delivery answer different questions.
Start with a specific decision. A marketing team deciding whether to explain its onboarding process needs different evidence from a team deciding whether a new page increases sales opportunities. The B2B website performance scorecard helps identify the weak area before choosing the method.
How much traffic does a B2B website need for A/B testing?
The requirement depends on the event being measured, its existing rate and the smallest worthwhile change. Eligible traffic to the affected experience determines how quickly the test can collect evidence. The whole website’s visit count can be misleading when only a small share reaches that experience.
Nielsen Norman Group’s testing guidance identifies low traffic as a limitation and recommends combining experiments with qualitative research. There is no universal monthly visitor threshold that makes every B2B test viable.
Consider a worked planning example. These are hypothetical numbers, unrelated to any Legency client. Suppose 2% of eligible visitors submit a valid enquiry. The team wants to detect an increase to 2.4%. That is a 20% relative improvement, equivalent to 0.4 percentage points.
A conventional fixed-sample, two-sided comparison of two proportions requires approximately 21,109 visitors per version, or 42,218 in total, using these assumptions:
- A 5% significance level, also called alpha.
- 80% statistical power to detect the specified change.
- Equal allocation between control and variant.
- Independent visitor-level observations and one binary conversion outcome per visitor.
- An unchanged experiment and a preselected final analysis.
This is a normal-approximation planning calculation, rounded up per version. It uses the pooled variance under the null and separate variances under the alternative. It excludes adjustments for repeated peeking, multiple comparisons and clustered observations.
At 1,000 eligible new visitors each week, collecting that total takes approximately 42.2 weeks before any additional observation period. The same assumptions for an increase from 2% to 3% require approximately 3,826 per version, or 7,652 in total. That takes approximately 7.7 weeks at the same traffic rate.
The larger effect is easier to detect. It also sets a less sensitive test. Choosing it because a deadline is approaching does not make a 50% relative improvement plausible.
Optimizely’s sample-planning guidance explains the relationship between effect size, variance and sample requirements. A platform using sequential or Bayesian methods needs its own planning assumptions. Record the method alongside the number.
Which visitors belong in the calculation?
Count people eligible to experience the change. For a demo-page test, that might be new visitors who reach the page. For a homepage proposition, it might be visitors entering a particular acquisition route. Define eligibility before assignment and keep it consistent across versions.
Avoid calculating the requirement from all sessions and then analysing unique visitors. One person can open several tabs or return repeatedly. A session-based analytics dashboard does not automatically provide the independent observations used in the example above.
Some B2B purchases involve colleagues from the same organisation. Their decisions can influence each other. If that matters to the experiment, account-level assignment and analysis may be appropriate. It needs a different sample calculation and usually fewer independent units than a visitor count suggests.
A publicity burst also changes the planning question. People arriving to admire a design may have different needs from people searching for a supplier. Record the source mix and ask whether the experiment’s audience represents the audience the page will serve after the burst ends.
What can a team improve before running a split test?
Start with failures that can be reproduced. A submit button that does nothing, an inaccessible error message or an enquiry that never reaches its destination can be checked directly. Repair the fault, repeat the task and document the result.
Give the test a realistic scope. Enter a long company name, use a mobile keyboard, trigger a validation error and return to the previous step. Check whether the entered information survives. Confirm that the confirmation message describes what actually happens next.
Then examine the information required to make a buying decision. Ask a representative buyer to find whether the service suits their company, identify relevant evidence and explain what they expect after contacting the business. Observe the task before asking for opinions about the design.
If several participants misunderstand the same phrase, there is a concrete research finding to investigate. Their confusion does not reveal a population conversion rate. Record the observation in its own terms, including who took part and the task they attempted.
For content gaps, review questions that sales repeatedly answers. A missing explanation of migration responsibilities, procurement requirements or ongoing ownership may warrant a useful page section. The business can make that decision without claiming a measured percentage lift.
How should a small team prioritise CRO work?
Use an evidence note for every proposed change. Keep the original observation separate from the proposed explanation. This avoids turning a plausible story into an established cause.
For example, an analytics report may show visitors reaching a form and leaving. That observation permits several explanations. They may have found the wrong service, wanted pricing, encountered an error or decided to return later. The drop-off alone cannot identify which explanation is correct.
Write a short priority record that another person can challenge:
- Audience and task: who is trying to do what, on which page?
- Observed obstacle: what happened, and where is the evidence?
- Commercial consequence: which useful outcome might be affected?
- Proposed change: what exactly will be different?
- Decision method: direct verification, buyer research, controlled experiment or an observed rollout?
- Cost and reversibility: what work is required, and how would the previous version be restored?
- Review condition: what result would justify keeping, revising or removing the change?
Prioritise verified failures affecting an important task. For competing hypotheses, consider the strength of the evidence and how much uncertainty the proposed research can remove. A numerical priority score can help organise a backlog, but its precision should not exceed the judgement behind it.
How do we measure progress with few enquiries?
Report counts beside rates. Two accepted enquiries from 100 eligible visitors and 20 from 1,000 share a rate, while providing different amounts of evidence. If the denominator is uncertain, say so before comparing periods.
Use separate measures for the stages a team can act on. A page visit shows exposure. A CTA click shows an attempted next step. A form submission shows completion of that task. Sales acceptance provides information about fit. An opportunity adds another commercial condition.
Agree what sales acceptance means. It might require an appropriate company, a relevant requirement and a viable buying conversation. Preserve rejected reasons such as job applications, supplier pitches and unsupported requests. A submission total cannot answer those questions by itself.
Give enquiries enough time to mature before comparing outcomes. An enquiry received yesterday has had less opportunity to become a meeting than one received last month. Use comparable follow-up windows and leave unresolved cases visible.
These definitions belong in a measurement and reporting process, together with the person responsible for reviewing quality. Repeatedly changing the qualification rule makes a trend difficult to interpret.
Can we use before-and-after results?
A before-and-after comparison can describe what happened following a change. Its interpretation depends on what else happened. Campaign changes, publicity, seasonality, service updates and differences in follow-up can influence the same result.
Record the release date, affected pages and concurrent activity. Compare complete periods and inspect the audience mix. Keep a note of outages and measurement changes. Describe the result as an observed association unless the design supports a causal conclusion.
For a low-volume site, a careful observed rollout can still inform a business decision. The report might say that completion rose after a form repair, with few submissions and a changed traffic mix. That is useful information when the next action is continued monitoring.
Avoid choosing the most favourable week after looking at every available window. Select the reporting window in advance, then include the periods that disappoint as well as those that improve.
A worksheet for the next CRO decision
Copy these prompts into the working document for one page:
Decision: The business needs to decide whether to ______.
Audience: The relevant visitors are ______. Exclusions are ______.
Evidence: The observed behaviour or failure is ______, collected on ______ using ______.
Alternative explanations: The same observation could also reflect ______.
Next research step: We will investigate ______ through ______.
Outcome: Success means ______, measured per ______ over ______.
Commercial check: We will also review ______ so that increased activity can be assessed for quality.
Method and limits: This is a ______ study. It can establish ______ and leaves ______ unresolved.
Owner and review date: ______ will review the evidence on ______.
A completed worksheet should reveal whether the next task is research, a repair or an experiment. If the team cannot define an outcome or eligible audience, spend the next work session on those definitions before building a variant.
Frequently asked questions
Is CRO worthwhile for an enterprise B2B site with limited traffic?
Yes, when the work addresses valuable buying tasks and the evidence suits the decision. Technical repairs, clearer information and buyer research remain useful. The available data determines how confidently their commercial impact can be quantified.
Should we replace qualified enquiries with CTA clicks to get a faster result?
Clicks can help diagnose a step in the process. Retain downstream quality checks when judging the change. A faster answer about clicks can leave the main commercial question unresolved.
How long should a low-traffic A/B test run?
Estimate the required sample using the chosen method, then divide the total by eligible traffic per period. Allow for normal business cycles and conversion delay. If the resulting duration is impractical, reconsider the research method or the question.
What should a B2B CRO agency provide first?
Ask for the affected audience, the evidence behind the proposed priority and the method for judging it. Our conversion rate optimisation service starts by connecting a website change to a defined buying task and an outcome the marketing team can review.
