Analytics, Measurement & CRO

Measure. A measurement plan names the business outcomes, the key events that represent them and the parameters each event carries. Data-quality checks confirm that events fire once, with the right values and with consent respected. Nothing downstream is more reliable than this layer.

Understand. Funnel analysis in GA4 or BigQuery shows where visitors leave; heatmaps and session recordings in Microsoft Clarity, form analytics, surveys and moderated user tests explain why. Each finding is recorded with its evidence and the audience it affects.

Test. Findings become hypotheses in a fixed format (because we observed this, we believe this change will move this metric for this audience) and are prioritised with ICE or PIE scoring. Pages with enough traffic get A/B tests; the rest get research-led changes measured against a stable baseline.

Learn. Every result, including losing and inconclusive ones, goes into a learning log that feeds the next round and the measurement plan itself. That feedback is why both disciplines run as one loop rather than two projects.

Conversion data now arrives incomplete by design. Visitors who decline cookies are not observed like those who accept, browsers limit identifiers and ad blockers remove some tags. Measurement has to be planned for that, not patched after the numbers drop.

Consent Mode v2, described in Google Analytics Help and Google Ads Help, passes the visitor's consent choice to Google tags and lets GA4 and Google Ads model part of the conversions they cannot observe. Under KVKK in Türkiye and GDPR in the EU, the consent banner must offer a real choice, and the measurement design starts from what the banner actually records in the browser, not from what a configuration screen claims.

Server-side tagging moves part of the collection to a server container under your own domain. It gives more control over which data reaches each vendor, but does not override a refusal and is not required for every site; we recommend it when governance, platform connections or performance justify its running cost.

For CRO this matters directly. If consent rates differ between variants, devices or traffic sources, observed conversion rates differ for reasons unrelated to the page. Experiment analysis states which data is observed and which is modelled, and compares variants on the same basis.

Research comes before testing. Analytics shows where the leak is; qualitative methods show why. The heatmaps and session recordings described in Microsoft Clarity documentation reveal rage clicks, dead clicks and scroll depth; form analytics shows which fields cause abandonment; short surveys and a few moderated user tests surface objections no click map can show.

An A/B test needs a hypothesis, a primary metric, a minimum detectable effect and a sample size calculated before launch, typically at 95 percent confidence and 80 percent statistical power. It runs for whole weeks, is not stopped when the dashboard first looks good, and is analysed once at the planned end. Guardrail metrics such as revenue per visitor or lead quality catch a variant that wins the primary metric while hurting the business.

Many sites lack the volume to detect realistic effects in a reasonable time, and the power calculation says so before anyone invests in a test. The alternative is research-led change: fix what research shows clearly, release changes one at a time where possible, compare against a stable baseline and a control segment, and read the result as directional evidence. Clear errors, such as a broken mobile form, need fixing, not testing. Your developers or website platform implement every change; we specify, prioritise and verify it, tracking included, and do not build websites.

A report is useful when it ends in a decision. Each programme runs three layers: a Looker Studio dashboard with one agreed definition per metric for day-to-day reading, a monthly decision report on what changed and what happens next, and a quarterly review of the measurement plan itself.

The monthly report covers data-quality status, conversion by channel and device on the agreed source of truth, the gap to platform-reported numbers, concluded experiments with their result and confidence, new research findings and next month's priorities. Search Console data sits beside GA4 so organic demand and on-site conversion are read together. Decisions and their reasoning go into the learning log.