Conversion Rate Optimization Checklist

The growth lead exports the last 28-90 days of sessions, conversion rate, and primary conversion event volume from GA4 for the target page or funnel. Confirm the conversion event is firing on actual submit, not on email-blur or button-click — mis-mapped events are the most common reason CRO tests show false lifts.

Pull the last two weeks of Hotjar, FullStory, or Microsoft Clarity recordings for the target URL. Look for rage clicks, dead clicks, scroll drop-off above the fold, and form-field abandonment patterns — these point to the friction the test should address.

Format: "Because we observed [evidence], we believe that [change] will cause [primary KPI] to [direction] for [audience]. We will know we are right when [metric threshold] is reached." The primary KPI must be a single conversion event — not a list. Secondary metrics (bounce, scroll depth, AOV) go in a separate field.

Use Optimizely's or VWO's sample-size calculator with the baseline rate, the minimum detectable effect (MDE) you care about, and 95% confidence / 80% power. If the math says you need 6+ weeks to reach significance on this page's traffic, the MDE is too small — pick a higher-traffic page or a bolder variant.

Copywriter drafts 3-5 headline options that lead with outcome, not feature. Pressure-test against the messaging house and ICP language pulled from sales-call transcripts. The control headline stays unchanged in variant A.

Designer mocks the variant in Figma with above-the-fold CTA, mid-content CTA, and sticky-footer CTA on mobile. Keep button copy action-oriented ("Get the demo" not "Submit"). UTM parameters on the destination URL must follow the team's source / medium / campaign convention.

Coordinate with RevOps on which fields the lead-routing rules actually need. Company-size and industry can often be enriched downstream by Clearbit or ZoomInfo rather than asked on the form. Inline validation on email and phone reduces submit errors.

Customer logos, a named testimonial with title and company, or a G2 / Capterra rating badge. Any quantitative testimonial claim ("we tripled pipeline") needs a substantiation file per FTC Endorsement Guides — link the source case study or attribution data.

Run the variant URL through PageSpeed Insights and confirm LCP under 2.5s, INP under 200ms, CLS under 0.1 on mobile. Page weight regressions from added images or video embeds will eat your conversion lift before the test even starts.

Marketing ops opens Google Tag Manager Preview, walks the variant flow end to end, and confirms the primary conversion event fires once on actual form submission — not on email-blur, page-scroll, or button mousedown. Check the DebugView in GA4 to confirm the event reaches the property with the correct parameters.

OneTrust, Cookiebot, or Didomi must gate the Meta Pixel, LinkedIn Insight Tag, and any non-essential analytics until the user accepts. Test from an EU IP via VPN — pre-consent firing on EU traffic is a GDPR violation regardless of where the company is based.

Slack DM approvals don't count as audit trail. Capture timestamped approval from legal (claim substantiation, testimonial usage rights) and brand (visual identity, tone of voice) in the project record before launch.

Optimizely, VWO, or Convert handles random assignment. Confirm the goal metric in the test platform matches the GA4 conversion event one-to-one. Note the launch timestamp — significance calculations depend on it.

SRM check: actual traffic split should be within ~2% of the configured 50/50. A larger gap usually means a redirect, a caching layer, or a bot-filter is biasing assignment — the test data is unreliable until that's fixed. Run a chi-square SRM test or use the test platform's built-in alert.

Share a snapshot of conversion rate, lift estimate, and confidence interval — but explicitly do not call the test until the pre-registered sample size is hit. Peeking and stopping early inflates false-positive rate; flag this in the readout if anyone asks to call it early.

Pull final results once the pre-registered sample size is hit. Report the lift, the 95% confidence interval, and the p-value. If the CI crosses zero, the test is inconclusive — don't roll out the variant on the strength of a directional read.

Promote the variant from the test platform into the CMS as the new control. Confirm the GA4 event still fires post-rollout and that the test platform's snippet is removed or archived to avoid stale variant code drag.

An inconclusive result usually means the MDE was too small for the traffic, or the variant was not bold enough. Revisit the hypothesis with the growth lead and pick a higher-leverage element (offer, pricing display, hero proof) for the next round rather than re-running the same change.

Record hypothesis, variant screenshots, sample size, runtime, outcome, and learnings in the team's Notion or Airtable experiment log. This is what prevents the team from re-running the same test in 9 months when the original owner has rotated off.