Numuno Digital /
A/B testing your website: what if traffic is low?
When your website produces few enquiries, comparing two versions can look like a way to find the answer. An experiment needs a specific question, reliable measurement and enough visitors in the journey being tested. This guide helps you decide what to request from a provider and when fixing an observed problem is a more useful first step.
By Numuno Digital, an independent web design and marketing studio in Marbella.
Start with a decision rather than two designs
Google describes A/B testing as showing variants simultaneously to randomly assigned groups. Replacing a page this week and comparing it with last week can provide observations, but does not isolate the change in the same way. Campaigns, the calendar and the people arriving may also have changed.
Write down the decision you want to make. For example, whether an explanation beside the form helps enquiries include useful context. That defines what changes and what stays consistent. Comparing entirely different designs can assess the package, but will not identify which individual element explains a difference.
Count traffic that actually enters the experiment
Gather visits to the page being tested rather than the whole site's traffic. Record the period, devices and acquisition sources. Enquiries collected through a social platform's own form are not automatically participants in your website experiment.
Define the outcome as well. A button click, a received enquiry and a suitable prospect are different measures. Check receipt and keep commercial qualification separate. Agree how duplicate enquiries will be handled and which information can be associated with a variant while respecting privacy preferences.
Use a hypothesis you can inspect
Illustrative example, not a client case: an Alicante service business receives enquiries that omit the work required. Version A keeps its current form. B retains the fields, price and offer but adds a short explanation of the information needed for a useful reply. The question is whether that guidance improves useful enquiries, not whether the page looks prettier.
Before testing, ask someone to follow each version and explain what they expect after submission. Fix an evident fault such as a field preventing completion. You do not need to keep a broken journey in a variant to prove that it harms the visitor.
Check whether the experiment can answer in time
There is no visitor count suitable for every site and test. Optimizely's calculator takes inputs including baseline conversion and the smallest change to detect. Smaller differences can require more data. A sample estimate for another page is not your own requirement.
Request an estimate using your data and the proposed method. How much eligible traffic exists, what improvement would justify the work and how long can you wait? If the plan is impractical, review content, observe the enquiry journey and resolve real customer questions first. That supports practical decisions without claiming a statistical experiment.
Check assignment and receipt before interpreting results
Ask for evidence of which version was shown and that both journeys can be completed. Follow a test enquiry on mobile, confirm its arrival and check its record. A click event alone does not confirm receipt of a message by the business.
Research on telemetry loss warns that missing measurements can distort experimental conclusions. Differences therefore require a measurement check too. If only one version records a successful submission, the report may be comparing tracking systems rather than customer responses.
Agree the rules before launch
Document the hypothesis, primary measure, allocation, planned period and analysis and stopping rules. Microsoft's experimentation guidance stresses hypotheses and caution with changes introduced during an experiment. Record incidents and modifications affecting the journey.
Do not decide solely because a version is ahead today. Analysis depends on the method: fixed-horizon tests and tools with sequential analysis have different rules. Ask which applies and how uncertainty is handled. Checking a technical fault is necessary; picking a favourable day to announce a winner is a different decision.
An inconclusive result still needs a next step
If the data cannot distinguish variants under the agreed criteria, the report should say so. This does not establish that they are identical or justify promising a conversion improvement. You may retain the clearer or easier-to-maintain option for design reasons, identifying that choice honestly.
Next steps may involve better data, a revised hypothesis or a more important problem. If enquiries arrive but do not fit, examine what people understood and which service they needed. More form clicks alone cannot correct an offer aimed at people unable to buy it.
What to request in your website proposal
Share the URL, page objective, available traffic sources and an enquiry summary without unnecessary personal information. Separate diagnosis, design changes and any experiment in the proposal, including access, tools, costs and responsibilities. Installing analytics is not the same as running a test: Google's A/B testing guidance specifies an external experimentation integration for Analytics.
Numuno Digital can assess your website journey and proposed improvements. Where an experiment makes sense, implementation and monitoring are scoped in the proposal. We also work online with Alicante businesses from Marbella. Identifying the missing decision does not require commissioning a complete redesign first.
Reviewed on 6 November 2026. The example and cover are illustrative. Publishing this guide does not launch an experiment or promise a conversion improvement.
Your checklist before starting.
- A defined question and business decision
- Actual page traffic and a recorded period
- An outcome separate from clicks and sales
- Variants and enquiry journey checked on mobile
- Agreed feasibility and analysis rules
- An incident and change record
- Reporting that can acknowledge an inconclusive result