How to measure whether your new landing page worked

Start with the question you want answered

You launch a new landing page. Revenue rises. During the same period, you increase ad spend, change the audience and introduce a bundle. How much of the improvement came from the page?

A launch date and a sales chart cannot answer that on their own. Agree the comparison before the work starts. Where traffic and implementation allow, run the existing page and the proposed page at the same time with randomly assigned visitors. Where that is impractical, be explicit about what the available evidence can establish.

This matters when you are deciding whether to keep the page, commission another project or commit to ongoing conversion rate optimisation. You need a report that helps you make that decision, including when the result is uncertain.

“Did revenue increase?” and “Did the new page cause additional purchases?” require different evidence.

Your store reports can tell you what happened to sales. A comparison with a credible control helps estimate what would have happened without the change. That second question is the basis of incrementality.

Write a specific question before launch. For example: “For eligible visitors arriving from this campaign, does the new destination increase completed purchases compared with the current destination?” Define eligibility using information available before the visitor encounters the changed experience. Selecting only visitors who clicked the new page’s buying button would leave out people whom the page failed to persuade.

If you change the offer as well as the page, name both in the question. You can evaluate the combined experience, but you cannot then claim the design alone produced the result.

Attribution tells you where credit went

Attribution assigns credit for a purchase or another action to marketing touchpoints. Google’s attribution documentation explains how that credit depends on the selected model.

Read Google Analytics attribution guidance

An attribution report can be useful for understanding recorded customer journeys. It does not, by itself, isolate the effect of replacing a landing page. If every campaign visitor now sees the new page, purchases associated with those campaigns will include customers who might have bought through the old page too.

For advertising, Google’s Conversion Lift explanation distinguishes attributed conversions from additional conversions estimated using an exposed group and an ad holdback group. That is a different comparison from a page test: it asks about the advertising, while a destination test asks about the experience after arrival.

Read Google Ads Conversion Lift guidance

Keep those questions separate in the brief. A successful page test does not demonstrate that every purchase attributed to the ad was caused by advertising.

A revenue increase can leave the page question unanswered

Consider this hypothetical before-and-after report. These are invented figures to illustrate the comparison, not a client result, experiment or forecast. Both periods have equal duration. Assume the same reporting rules and that all listed revenue and spend relate to the same defined campaign group.

Revenue increased by $10,000, or 50%. Spend and eligible session volume also rose by 50%. The reported revenue-to-spend ratio and purchase rate stayed unchanged. Assume average order value stayed unchanged too.

The business sold more during the second period. The table provides no estimate of how much the new page contributed. If the audience or offer changed, the page could have helped, hurt or made little difference while other changes affected the totals.

Even if the purchase rate had risen, the missing comparison would remain. A warmer audience or stronger discount could have changed that rate. Dividing by traffic removes the volume difference; it does not make the two audiences equivalent.

Compare destinations concurrently when feasible

For a page-only question, a practical design is to randomly assign eligible arriving visitors to the current destination or the new destination during the same period. Keep their assignment consistent across return visits where the implementation permits. Document identity and consent limitations rather than assuming every person can be recognised across browsers and devices.

Both groups should face the same offer and buying conditions. If a promotion starts midway through the test, apply it consistently and record the change. Randomisation helps distribute outside influences between groups; it cannot guarantee perfectly balanced groups or protect a broken implementation. The NIST guide to experimental design explains the role of randomisation in managing nuisance variables.

Read the NIST experimental design guide

Two separately optimised ad campaigns, each pointing to a different page, can deliver different audiences. Equal budgets alone do not establish a randomised page comparison. Ask where assignment occurs and whether the platform can preferentially send different people to each version.

Changing spend during a properly implemented concurrent page test does not automatically invalidate it. What matters is whether both versions continue to receive randomly assigned visitors from the same eligible traffic pool. The result describes that traffic mix and period. A later expansion into a new market or much colder audience deserves a fresh check.

Choose the outcome before looking at the winner

For a purchase-focused page, completed purchases are usually closer to the business question than button clicks. Define exactly what counts, the reporting window and the denominator. A visitor-based experiment and Shopify’s session-based conversion report can describe different quantities; do not silently substitute one for the other.

Decide how you will handle repeat visits, delayed purchases, cancellations and refunds. If the page changes discounting or basket composition, include revenue or contribution per assigned visitor where the data supports it. More orders may come with less money retained per order.

Use add-to-cart and checkout progression to investigate why an outcome moved. Avoid promoting whichever intermediate metric improved to the primary result after seeing the data. Optimizely’s guidance on primary and secondary metrics explains why the distinction matters.

Read Optimizely’s metric guidance

Include checks for harm, such as buying errors or slower page loads. Agree what would justify stopping for customer protection. That decision can be necessary before the test has enough evidence to identify a commercial winner.

Plan for the traffic you can actually test

There is no useful universal rule that every store needs the same number of visitors or that every experiment should end after the same number of weeks. Plan around the eligible audience, baseline outcome, effect worth detecting and statistical method. For revenue or contribution metrics, variability matters too.

A sample-size calculation is a planning tool. Its inputs and assumptions belong in the measurement plan. Optimizely’s sample-size calculator illustrates how baseline conversion, minimum detectable effect and the significance setting change the requirement. Use a method appropriate to your experiment, including its power assumptions, rather than copying another store’s traffic target.

Explore Optimizely’s sample-size calculator

Ask whether the planned experiment could detect an improvement small enough to matter commercially. If only a very large change is detectable in the available time, a modest but valuable benefit may remain unresolved.

Agree how results will be reviewed. Repeatedly checking an ordinary fixed-horizon test and stopping at the first favourable result can distort inference. A sequential method can support interim decisions when used according to its rules. Microsoft’s during-experiment guidance discusses the need to account for early peeking and repeated testing.

Read Microsoft’s experiment monitoring guidance

A report should show the estimated effect and its uncertainty, alongside data-quality checks and the commercial decision. An inconclusive result does not establish that the two pages are equivalent.

Useful work when traffic is limited

When a store has limited traffic, agree which paid work can still be judged directly: a repaired buying action, a verified tracking setup or a customer-research report with specific findings. The Shopify product-page checklist covers practical checks on the buying journey. Completing those checks does not quantify incremental revenue.

Review the Shopify product-page checklist

If a randomised commercial test would take too long to inform the decision, agree a bounded implementation project with observable acceptance criteria. Keep reporting business outcomes, but describe before-and-after changes as observational. Record competing explanations rather than converting every post-launch increase into a claim about the work.

A specialist may propose a different causal design. Ask what makes the comparison credible, which assumptions it requires and what could break those assumptions. Changing the method does not remove the need for a defensible comparison.

Agree measurement before buying a retainer

An ongoing CRO programme needs enough useful work and enough capacity to act on what is learned. Before committing, ask the agency to put the following in the proposal or measurement brief.

1. The first decision the work should inform. Specify the audience, proposed change and current comparison. State whether you are testing a page, an offer or a combined experience.
2. The measurement definitions. Name the primary outcome, denominator, observation window and source of truth. Explain how order data, refunds and experiment assignments will be reconciled.
3. The feasibility assessment. Show the eligible traffic available, the effect worth detecting and the method used to estimate a realistic evaluation period. State what happens if the traffic forecast is wrong.
4. The delivery responsibilities. Identify who implements assignment and tracking, checks the buying journey, records outside changes and reviews results. Ask how they will detect missing purchases, duplicate events or an unexpected split between variants. Confirm access to the report and experiment history.
5. The decision rules. Agree how to handle harm, a credible benefit, an inconclusive result and unreliable data. Include how the team will decide to roll back, iterate, keep the experience or investigate further.
6. The commercial review. Separate completed work from measured impact. Set a point to review whether the next period of work is justified by the remaining problems, feasible learning and total costs.

If fees depend on performance, ask exactly which revenue qualifies, what comparison establishes the gain, and how discounts, refunds, media changes and later reporting corrections are handled. Sending all traffic to a new page does not establish that all revenue through it is additional. Resolve that distinction before using the report to calculate a fee.

For a store with little testable traffic and a known implementation problem, a focused project may be the sensible starting point. A retainer becomes easier to evaluate when there is a continuing set of important questions, reliable measurement and the ability to implement changes. No agency can responsibly promise that every monthly test will produce a detectable winner.

Ask for the comparison behind the result

When reviewing a CRO report, ask: “What would we expect to have happened without this change, and what evidence supports that comparison?”

The answer should make the next decision clearer. It might support keeping the page, justify a further test or show that the present data cannot settle the question. If the goal is profitable advertising scale, a page result is one input; acquisition costs, margins and the audience reached at higher spend still matter.

For the broader sequence of research and implementation, use the CRO roadmap. For a discussion about measurement on your own store, bring the current destination, recent traffic mix and known tracking gaps to a strategy call.

Read the CRO roadmap

Discuss measurement on your store