Web Analytics Checklist

Crawl the site with Screaming Frog's Custom Search for the GA4 measurement ID (G-XXXXXXX) and confirm the tag fires on every template — including subdomains, blog, checkout, and gated thank-you pages. Common gotcha: a marketing site migration to a new CMS where the tag was never re-added to the new templates.

Walk a representative session through Preview mode: home, key landing page, form submit, thank-you page. Confirm only one GA4 config tag fires per page (duplicate tags from legacy UA migrations are common). Cross-check in GA4 DebugView.

The CMP (OneTrust, Cookiebot, Didomi) must gate analytics_storage and ad_storage before any pixel — GA4, Meta, LinkedIn — fires. Test in incognito with a fresh consent state. A banner that promises 'we don't track without consent' while Meta Pixel loads pre-consent is a GDPR exposure.

Open a ticket with web engineering to wrap the offending tags in CMP consent triggers. Do not proceed with the rest of the audit until the leak is closed — downstream analytics data collected pre-consent should not be relied on for decisions.

Lock down source / medium / campaign / content / term values in a shared link-builder (Sheets or a tool like UTM.io). Drift across teams is the most common reason channel reports become uncomparable quarter-over-quarter.

List the events that count as Key Events (formerly conversions) in GA4 — typically demo_request, signup_complete, purchase — and the micro-conversions that feed nurture (newsletter_signup, content_download, video_50_complete). Note which downstream system consumes each: HubSpot, Salesforce, Meta CAPI.

A classic GA4 misconfiguration: form_submit firing on field blur or button click rather than the actual submit/thank-you. Inflates reported conversions by 3-10x and distorts CAC. Test by abandoning the form mid-fill and confirming no event fires.

Looker Studio or in-platform Explorations: sessions, key events, conversion rate, and CAC by source/medium with a 28-day window. Use the documented UTM convention from the prior section as the source-medium dimension — anything 'not set' or '(other)' goes on a tagging-fixes followup list.

Weekly digest to demand-gen, content, and the CMO with WoW deltas on key events, top landing pages, and channel pacing vs. plan. Keep it to one screen — long auto-reports get ignored within a month.

Layer a reverse-IP / visitor-identification source (Clearbit Reveal, 6sense, Demandbase) onto GA4 audiences. Compare on-ICP vs. off-ICP behavior on key pages — pricing, demo, case studies. Off-ICP traffic dominating a high-intent page is a paid-targeting problem, not a content problem.

GA4 Path Exploration starting from demo_request, going backwards. Identify the top 3 entry points and the typical 3-5 page sequence. Use this to prioritize CRO tests in the next section — optimize the pages people actually traverse, not the ones leadership talks about.

Hotjar, Microsoft Clarity, or FullStory: scroll depth, click maps, and 5-10 session recordings per page. Watch for dead clicks on non-interactive elements (people clicking on a stat that looks like a link) — those are CRO opportunities.

Score each candidate test on Potential, Importance, Ease (PIE) or Impact, Confidence, Effort (ICE). Anchor scoring in the path-exploration findings — tests on pages with high traffic and meaningful drop-off rank above 'redesign the footer.'

Pre-calculate required sample size for a minimum detectable effect of 10-15% at 95% confidence. Most B2B sites underpower tests and call winners early. Forward the GA4 client_id to the testing tool so results can be segmented post-hoc.

Hold the test until the pre-calculated sample size is reached and at least one full business cycle (typically 2 weeks) has elapsed. Calling a winner on day 3 because variant B is 'up 20%' is the most common CRO mistake — early peeking inflates false positives.

Hand the winning variant to web engineering for permanent implementation in the CMS. Confirm the post-launch conversion rate matches the test result over the following 30 days; regression to control is common when the test environment differed from production.