Landing Page A/B Testing Checklist

A landing page test can easily teach the wrong lesson. Traffic shifts partway through, tracking breaks on one variant, or someone stops the test the day the dashboard looks good, and a random swing gets rolled out as a win.

Use this checklist to run tests that end in decisions you can defend. It applies to SaaS pages, where the goal is qualified demos or activated trials, and to e-commerce pages, where the goal is orders and revenue.

Hypothesis quality

A strong hypothesis links evidence to a change, the change to an expected outcome, and the outcome to a metric. Without that chain, a result can't tell you why something worked, so the lesson won't carry to the next page.

Checklist

0/6

Start from evidenceUse session recordings, form analytics, sales call notes, support tickets, reviews, or surveys. A hunch or a competitor's page can spark an idea, but it isn't evidence.
Test one idea per variantA variant can include several edits if they express one idea, such as clearer pricing. Mixing ideas makes it impossible to know which one moved the metric.
Name the audienceSay which visitors the change is for: a traffic source, device type, or buying stage. A change written for paid search visitors may not suit returning customers.
Predict the outcome and the reasonWrite down what you expect to happen and why. If the test fails, the reason points to the assumption to question next.
Make the change big enough to detectSmaller effects need more traffic to detect. If your traffic can't support the test, try a bolder change or a higher-traffic page.
Confirm a result would change a decisionWrite down what you'll do if the variant wins, loses, or shows no clear difference. If every outcome leads to the same action, skip the test.

Common test types by business model

The same page element does different work in each model. Use the table to generate hypotheses for your own funnel, not as a list of changes known to work. For wider context, see SaaS growth marketing and e-commerce growth marketing.

Common landing page test types
Test area	SaaS	E-commerce
Offer and call to action	Demo request versus free trial, or a product tour first	Bundle versus single item, or a free shipping threshold
Message match	Headline that echoes the ad's problem versus a product-led headline	Headline that echoes the ad's product versus a brand-led headline
Proof and risk reduction	Security details, integration lists, or setup steps	Reviews near the buy button, returns policy, or delivery dates
Form or checkout friction	Fewer form fields, or questions that route leads by fit	Guest checkout, express payment, or fewer checkout steps
Pricing presentation	Plan comparison layout, or annual versus monthly billing first	Price per unit, subscription versus one-time, or bundle savings
Page structure	Short page with one path versus a long page that handles objections	Product details first versus lifestyle imagery first

Measurement plan

Decide what success means before any data arrives. A test with several equal metrics is likely to show something moving by chance, which makes it easy to declare a win that isn't real.

Revenue per visitor

Revenue from a variant's visitors ÷ Visitors assigned to that variant

For e-commerce, this combines conversion rate and order value, so a variant that adds orders at a lower value only wins if revenue rises. If margins vary widely by product, consider margin per visitor instead.

Checklist

0/5

Choose one primary metric tied to revenueFor SaaS, use qualified demos booked or trials activated, not raw form fills. For e-commerce, use orders or revenue per visitor, not add-to-cart rate. This metric decides the test.
Define “qualified” and “activated” in writingUse an exact rule, such as a demo accepted by sales or a trial that completes a key setup step. Changing the definition after results arrive invalidates the test.
Pass the variant to your CRM or order dataStore the variant on the lead record in HubSpot or Salesforce, or on the order in Shopify. Lead quality and refunds take time to appear, so you'll need that link later.
Set guardrail metricsGuardrails are metrics that must not get worse, such as sales-accepted lead rate, average order value, refund rate, or page load time. A variant that breaks a guardrail is not a win.
Write the decision ruleState the smallest effect worth acting on, the statistical confidence you require, and what happens after a flat or negative result. Agreeing now prevents debate over ambiguous numbers later.

Before the test

Checklist

0/8

Estimate sample size and durationUse a sample size calculator with your baseline conversion rate, the smallest effect worth detecting, and expected traffic. If the duration is impractical, change the test instead of hoping for a big effect.
Plan to run in full weekly cyclesWeekday and weekend visitors can behave differently, so cover whole weeks. For SaaS, also watch for events like quarter-end sales pushes that change how demos are followed up.
Define which traffic enters the testList the campaigns, channels, and audiences included. Results only apply to that mix, so a page tested on paid search traffic may perform differently for email or organic visitors.
QA every variant across devices and browsersCheck the devices, screen sizes, and browsers in your analytics data. Test forms, checkout, payment options, and load speed. A layout bug in one browser can decide a test on its own.
Verify tracking end to endSubmit a test lead or place a test order on each variant. Confirm that primary and guardrail events fire once, carry the variant ID, and appear in GA4 and your CRM or store.
Check for flickerClient-side testing tools can briefly show the original page before the variant loads, which can change behavior by itself. Check on slower connections and fix any visible flash.
Exclude internal and bot trafficFilter out your team, QA sessions, and known bots, so their visits don't mix with buyer behavior.
Get the plan signed offRecord the hypothesis, variants, audience, metrics, sample size, duration, and decision rule. Share it with whoever will act on the result before launch.

During the test

Your main job now is protecting the test's integrity. Watch for problems, not for winners.

Checklist

0/6

Stick to the planned stopping pointChecking daily and stopping when results look significant raises the chance of a false win. If you need early looks, use a sequential testing method designed for them.
Monitor for sample ratio mismatchIf visitor counts don't match the split you planned, something may be broken, such as a failing redirect, a bot filter, or caching. Check the counts with a chi-square test, and pause analysis until you find the cause.
Keep traffic sources stableAvoid changing budgets, targeting, ads, or offers for campaigns that feed the test. If a change can't wait, log the date and decide whether the test needs a restart.
Leave the variants aloneEditing copy or design mid-test mixes two experiments into one. If a fix is essential, stop, correct it, and start a fresh test.
Stop early only for real harmEnd a test early if a variant breaks something, such as a form that won't submit or failed payments. A disappointing trend is not a reason to stop.
Log outside eventsNote promotions, email sends from Klaviyo or HubSpot, outages, pricing changes, and press coverage. You'll need this log to interpret the results.

Hypothetical sample ratio check

Hypothetical numbers

Planned split: 50% control and 50% variant
Visitors recorded: 10,000 in control and 9,200 in the variant
Expected with an even split: 9,600 in each
Chi-square statistic: (400² ÷ 9,600) + (400² ÷ 9,600) ≈ 33.3
An imbalance this large is very unlikely to come from random assignment, so find the cause before reading results

Many testing tools and online calculators can run this check for you.

After the test

Checklist

0/7

Confirm data quality before reading resultsRerun the sample ratio check, confirm the test reached its planned sample, and review the event log. If data quality fails, the result isn't usable, however good it looks.
Apply the decision rule you agreed onCompare the result with the minimum effect and confidence set before launch. A result that falls short is inconclusive, so record it that way rather than as a small win.
Check guardrails and downstream qualityFor SaaS, follow test leads through qualification and pipeline in your CRM. For e-commerce, check refunds and returns. A variant can win on the page and lose later in the funnel.
Look for novelty and seasonality effectsReturning visitors may react to anything new, so an early lift or dip can fade. Sales, holidays, and launches also shift behavior. Check whether the effect held across the whole test.
Treat segment results with cautionSlicing results by device, source, or region many ways is likely to surface differences by chance. Act on a segment only if you planned it before launch. Otherwise, make it the next hypothesis.
Document the result and the decisionRecord the hypothesis, variant screenshots, dates, traffic sources, sample sizes, results, guardrails, the decision, and what you learned. Include flat and losing tests so no one repeats them.
Roll out winners and keep measuringShip the winner to all traffic and keep watching the primary metric and guardrails. Effects don't always hold over time, so re-test important wins when traffic mix, offers, or pricing change.

How to use this checklist

Copy the items into your test plan template and review them at three points: when a hypothesis is approved, the day before launch, and before anyone reads results. Not every item fits every test. When you skip one, write down why.

If tests rarely reach a clear result, the constraint may be traffic or tracking rather than ideas. For help building a research-backed backlog, see our approach to conversion rate optimization. For page builds and test variants, see landing pages.

Landing Page Experiment Checklist

Key takeaways

Hypothesis quality

Common test types by business model

Measurement plan

Before the test

During the test

After the test

How to use this checklist

Attribution Readiness Guide

Paid Search Account Audit Checklist

Find the constraint before spending more to scale it.

Key takeaways

Hypothesis quality

Common test types by business model

Measurement plan

Before the test

During the test

After the test

How to use this checklist

Related reading.

Attribution Readiness Guide

Paid Search Account Audit Checklist

Find the constraint before spending more to scale it.