CRO and Conversion Strategy

Landing page testing when your site has little traffic

 ·  13 min read

Hands comparing two printed landing page layouts beside a navy notebook and red pen
In this article

Landing page testing with low traffic starts with a choice: can a split test collect enough evidence to guide your decision? If it cannot, fix confirmed faults, watch likely buyers use the page and make a measured change you can reverse. A small sample can reveal a problem without proving a conversion lift.

A founder with a few enquiries each month still needs to improve the page. Waiting for a large audience leaves real problems in place. Declaring a winner after a handful of signups creates a different problem: the next budget decision rests on a result that may disappear.

We separate three jobs: checking that the page works, understanding why buyers hesitate and measuring whether a change improves an outcome. Each job needs different evidence. You can complete the first two while you decide whether the third deserves a live experiment.

Choose the decision before the test

Write down the decision that a result would change. “Should we explain the implementation process before asking for a call?” gives you a useful question. “Can we increase conversion?” leaves the page, audience and offer open to constant changes.

Choose one primary outcome that matches the commercial decision. For a service page, that might mean valid enquiries from businesses you can help. For a product page, it might mean purchases. A click on a button can help diagnose the path, but it does not establish that a buyer completed it.

Keep the measurement close enough to the change to explain the result. A test of form wording should track successful submissions as well as starts. A test of delivery information should follow the purchase path. If your sales process takes time, let each group reach the same follow-up window before comparing qualification or revenue.

Write the limits too. You may accept a little more form effort if it reduces unsuitable enquiries. You may reject a version that attracts more orders but creates more cancellations. Define those trade-offs before a dashboard shows a preferred answer.

Our guide to website traffic without leads covers the broader diagnosis. This decision comes later, once you can name the page problem and the action you want to compare.

Count the visitors who can enter the test

Use traffic to the actual landing page, after your eligibility rules. Total website visits can include blog readers, existing customers and people who never see the offer. Those visits cannot support a test of a page they did not enter.

Keep the numerator, denominator and dates consistent. Count purchases against the corresponding eligible audience during the same period. Do not divide annual orders by one month’s traffic. Do not divide units sold by visitors and call the result the share of visitors who bought: one buyer can purchase several units.

Record whether your tool assigns a version by visitor, account or session. Use the same unit in planning and analysis. Repeat visits from one person do not automatically provide the same information as visits from different people. Check how the tool keeps a returning person in a version and records their eventual conversion.

Build a small baseline from your own records:

Check Record before launch
Eligible audience The page, locations and audience rules
Allocation unit Visitor, account or session
Primary result The action and qualification rule
Observation window Matching start, end and follow-up dates
Exclusions Test visits, known spam and internal activity

Use our conversion rate calculator to check the arithmetic. The calculator cannot tell you whether you chose comparable inputs. That check belongs in the written test plan.

Check whether a split test can finish

Estimate the sample before building a variation. Your starting conversion rate, the improvement you need to detect and the statistical method affect the required audience. A smaller detectable difference generally needs more observations. No single monthly traffic threshold makes every landing-page test useful.

Optimizely’s minimum detectable effect guidance explains how the difference you want to detect shapes experiment design. Choose a difference that would change a business decision, then check whether your expected traffic can support it.

Suppose a page currently converts 2% of eligible visitors. A rise to 2.4% means an increase of 0.4 percentage points, or 20% relative to the starting rate. Those are two ways to describe the same hypothetical change. Neither says how many people you need; your sample calculation must include the chosen method and error settings.

Translate the required sample into time. As a separate planning illustration, assume your calculation requires 4,000 eligible visitors in each of two equally sized groups. At 200 eligible visitors a week in total, collecting 8,000 visitors takes about 40 weeks. The 4,000 figure is an assumption for this calendar example, not a sample-size recommendation for the 2% scenario.

That timetable may outlast the offer, campaign or decision. Choose a larger, evidence-based change, improve the page through research, or wait for a suitable traffic opportunity. Do not shrink the required sample to make a launch date fit.

Use a calculator that matches your experimentation tool. Optimizely’s sample-size calculator uses its own Stats Engine. Apply its estimate to that method. Other testing platforms can use different stopping rules.

A researcher points out a product-page overlay that blocks the buyer’s button.

Watch likely buyers try the page

When a statistical comparison would take too long, observe people who resemble the buyers you want. Give them a realistic task: decide whether the service fits their business, find the delivery terms or start an enquiry. Ask them to explain what they understand and what they still need.

The GOV.UK guide to moderated usability testing describes observing participants as they attempt specific tasks. Use that method to investigate confusion and obstacles. A successful session does not estimate the conversion rate of your entire audience.

Avoid teaching the page while you test it. “Where would you look for the delivery date?” leaves room for the person to choose. “Did you notice the delivery date beside the button?” points them toward your intended answer. Record the route they take before explaining anything.

Watch for the difference between preference and difficulty. Someone may dislike a colour while still finding the price and completing the task. Another person may praise the design but misunderstand what they will receive. Prioritize the misunderstanding that blocks a decision.

Compare observations with real enquiry questions, support messages and reasons people declined. Keep personal details out of shared notes. A repeated question can support a copy change, but treat it as a clue about the people you observed. Do not turn three interviews into a claim that most visitors share the same concern.

Make a short record for each finding: the task, the point of confusion, the exact page element and the proposed repair. This gives the next version a reason to exist beyond someone’s taste.

Repair faults before comparing designs

Fix a broken button, an unusable mobile form or an incorrect price without treating the fault as a valuable control condition. First reproduce the problem, then verify the repair on the devices and paths your buyers use. Track the repair separately from any later experiment.

Check the full path after the click. A page can show a thank-you message while the enquiry never reaches your inbox or CRM. Compare a controlled test submission with its saved record and destination. Mark test records so they never inflate your lead count.

Our lead-tracking cleanup guide explains that check in more detail. Keep it outside the experiment’s primary result until you know the tracking works in both versions.

Check mobile layout with a real keyboard open. Confirm that the field label remains visible, error messages identify the problem and the submit control stays reachable. Read the page at a narrow width and with larger text. A tidy desktop screenshot does not establish that someone can complete the mobile task.

You can use the Website Lead Leak Checker as a starting point for page checks. Follow a reported issue through the actual visitor path before changing the design. A scan does not replace a completed test enquiry or a buyer conversation.

Then save a working baseline. Keep a copy of the page and record its release time, tracking setup and active campaigns. If the next version causes trouble, you need a known version to restore and a clear period to inspect.

Build one change around a real objection

A low-traffic page needs a test worth the available audience. Choose an objection that affects the buying decision: unclear scope, uncertain delivery, weak evidence or an unexplained commitment. A cosmetic change can use the same scarce audience while answering a much smaller question.

Turn the finding into a specific hypothesis. For example: “Buyers cannot tell what the first month includes. Showing the actual steps before the enquiry form should help suitable businesses decide whether to contact us.” This is a proposed explanation, not a promised improvement.

Change the relevant experience enough to address that objection. You may need a heading, a short process section and a clearer button together. If you test them as one package, your result applies to the package. It cannot tell you which individual element caused the difference.

Keep unrelated changes out. A new offer, discount, traffic source and page design introduced together make the result hard to explain. Document unavoidable changes, such as a campaign ending or a product going out of stock, and review whether the comparison still answers the original question.

Use one challenger against the current page unless your sample plan supports more. Splitting a small audience across several versions reduces the visitors available to each comparison. Save other ideas with the evidence behind them for a later test.

A designer compares two product pages that differ by one reassurance symbol beneath a customer’s delivery concern.

Set the rules before you launch

Record the page versions, audience, primary result, allocation method and decision rule together. Include the expected collection period and any delay between the visit and the outcome. Confirm that your experiment tool measures the outcome you intend to use.

Optimizely’s experiment setup documentation describes random traffic allocation and a primary metric. Follow the setup and analysis rules of your chosen system. A fixed-sample test and a system that supports continuous monitoring do not necessarily use the same stopping procedure.

Keep collecting safety signals even while you wait for the primary result. A failed checkout or a form that loses enquiries calls for investigation. Write down the conditions that justify stopping for harm, plus the conditions that allow a statistical decision. This prevents a convenient early fluctuation from becoming the only reason you stopped.

Check the exposure record too. Both versions need working measurement, and the audience allocation must follow the plan. If one version fails to load or visitors switch between versions unexpectedly, repair the setup before interpreting the result.

If you buy more visits, keep their intent relevant to the decision. A burst of cheap traffic from a new audience can change who you measure. Discuss the audience and landing-page match with your paid ads team before using spend to shorten a test. More visits help only when they belong to the audience the decision concerns.

Read small results without inventing a winner

Suppose two versions each receive 100 eligible visitors. One records four enquiries and the other records six. The observed rates are 4% and 6%, so the second version shows a 50% relative increase in this hypothetical sample. That calculation alone does not establish a reliable improvement.

Keep the counts beside the percentages. A change of two enquiries looks different from a large result described only as “50% higher.” Review the uncertainty reported by the test method, the sample plan and the quality of those enquiries before deciding what the result supports.

NIST’s guidance on confidence intervals for proportions explains methods for expressing uncertainty around a measured proportion. A point estimate tells you what happened in the sample. It does not remove the uncertainty about performance beyond that sample.

Use clear outcome labels: evidence supports the planned change, evidence supports keeping the current page, or the result remains inconclusive. An inconclusive comparison does not prove that the pages perform equally. It may mean the test cannot separate the difference from the noise at the settings you chose.

Keep subgroup checks in proportion. A small total becomes even smaller when you split it by device, source and location. Use a surprising segment to form the next question unless your original design supports that separate conclusion.

A researcher weighs two small samples while a stack of additional observations sits on the table.

Make the next decision easy to revisit

If you cannot support a useful split test, choose a reversible improvement from the evidence you do have. Record why you chose it and what would make you undo it. A before-and-after comparison can help monitor the change, but shifting traffic or seasonality can also move the result.

Keep a short decision log with the old page, new page, release date, audience changes and observed outcomes. Separate a confirmed technical repair from a research-led design choice and a measured experimental result. That distinction keeps later reports honest.

Review the page when the evidence changes: a new offer, a different audience or enough eligible traffic for a planned comparison. You do not need to retest every sentence each week. Spend the next round on the question that would change the next commercial decision.

If your team needs help choosing that question, our website conversion work starts with the page and the action you need it to support. Bring the traffic counts, recent enquiries and the change you are considering. Those inputs help define a useful scope before you pay for another design.

FAQ

How much traffic do I need for landing page testing?

You need enough eligible visitors for the outcome and difference you want to measure. Calculate the required sample using your baseline, the minimum detectable effect and your tool’s statistical method. Then translate that sample into time using traffic to the actual page. A fixed monthly threshold cannot answer that question for every business.

Can I improve a landing page without an A/B test?

Yes. Repair confirmed faults, check the enquiry or checkout path and watch likely buyers attempt realistic tasks. Use those findings to make changes you can reverse. Record the evidence and monitor the result. Describe the change as a repair or research-led decision unless a suitable experiment supports a conversion claim.

Should I use button clicks when purchases are rare?

Button clicks can help explain where people move or stop, but they do not prove that purchases or qualified enquiries increased. Keep the business outcome as a separate measure. If you use a more frequent action as the primary test metric, state that limited goal and check downstream outcomes before claiming commercial success.

How long should I run a low-traffic landing page test?

Estimate the duration before launch from the required sample and eligible traffic, then allow time for delayed outcomes. Follow your tool’s stopping rules and account for the business cycles relevant to the decision. If the required period outlasts the offer or campaign, reconsider the experiment and its deadline.

Does a higher conversion rate always justify the new page?

No. Check whether the conversions match the business goal and whether the change creates costs elsewhere. A version that attracts more unsuitable enquiries, cancellations or support work may not improve the business. Define those checks before launch, keep counts beside rates and compare outcomes over equivalent follow-up windows.

Free website and CRO audit

Find where your website loses enquiries

We check where visitors drop off and identify page changes to test, then send the fixes in priority order.

Your data is safe. We never share details without permission.

Free growth audit

Find out what's costing you leads

We check your site, ads and tracking and send the fixes that matter most within 24 hours. Your first 15 days are free on a new plan.