Analytics & Data Strategy

Every engagement begins with a measurement plan: a short document listing the business questions, the KPIs that answer them, the events and parameters that produce those KPIs, where each event fires, and who owns it. It is written with marketing, sales and development, because an event engineering cannot implement is a dead row.

The plan separates key events (the few actions that define success: a qualified lead, a booking, a purchase) from supporting events, and fixes naming conventions and value rules once so GA4, Google Ads and Meta receive consistent definitions.

We implement GA4 through Google Tag Manager with a documented data layer, not with tags hard-coded in templates. The data layer specification tells developers which object to push at which moment (page metadata, user state, transaction data); Tag Manager turns those pushes into GA4 events through triggers anyone can read.

Event and key-event design follows the plan: recommended GA4 names where they exist (generate_lead, purchase, sign_up), custom events only for what Google does not define, and parameters registered as custom dimensions before a report needs them. Each key event gets a counting method, a value rule and a test case, so releases are verified in DebugView rather than discovered as gaps later.

Server-side Google Tag Manager moves tag execution from the browser to a server container you control, hosted on Stape or on your own cloud project under a first-party subdomain. The browser sends one request; the server forwards clean events to GA4, Google Ads, Meta and other destinations. Pages load fewer third-party scripts and you decide which fields each vendor receives.

It does not bypass consent: a server container receives only what the browser is permitted to send. We implement it where volume or privacy posture justifies the running cost; for a small site a well-built client-side container is often the right answer.

Consent Mode is Google's mechanism for passing a visitor's consent choices to its tags. Version 2 added the ad_user_data and ad_personalization signals, which Google requires for advertising features aimed at users in the European Economic Area. According to Google Analytics documentation and Google Ads Help, tags adjust their behaviour to the consent state, and in the advanced implementation cookieless pings let Google model conversions that consented data cannot observe.

We wire Consent Mode to your consent platform so the default state loads before any tag and the signals reach both the web and the server container. For businesses in Türkiye, KVKK and the Personal Data Protection Authority's cookie guidance apply alongside GDPR for European visitors. We implement and document the technical side; your counsel decides the legal basis and banner text. We do not provide legal advice.

Google Ads imports key events from GA4 or receives its own conversion tag through Tag Manager; we choose one source per conversion to avoid double counting, enable enhanced conversions where hashed first-party data is available with consent, and align value rules with the measurement plan, because smart bidding learns from that data.

Meta's Conversions API sends events from the server container alongside the browser pixel; Meta Business Help documents deduplication by event_id, so each lead or purchase is counted once. The same pattern applies to TikTok Events API, each destination receiving only the parameters it needs.

GA4 exports raw, event-level data to BigQuery at no cost for standard properties, the only place for unsampled analysis with long retention. We set up the export, design a schema with sessionised and conversion tables, and write scheduled queries that produce the plan's metrics, joined with CRM, order and ad-cost data.

Not every business needs BigQuery; we recommend it when the GA4 interface cannot answer your questions. Looker Studio sits on top: an executive dashboard with a few defined metrics, channel dashboards for operators, and a metric glossary so every number has one definition.

Traffic and demand now arrive through AI assistants as well as search and social, and the measurement plan has to keep five different signals apart. Human referral sessions are visits that arrive from an assistant's link and behave like any other referral; they are measured in GA4 like other channels, provided the referrer is recognised and grouped. External bot requests are fetches by AI crawlers and user-triggered fetchers; they live in server or CDN logs, never in GA4, and are counted by user agent and verified where the platform documents its ranges. Sampled model visibility is how often assistants mention or cite the brand in a fixed prompt set; it is a sample with a denominator, not a traffic figure. Owned-agent tasks are actions performed by automations you run yourself, which need their own event stream and identifiers. Verified business outcomes are leads, bookings and revenue confirmed in your systems.

Two rules follow. Missing data is not zero: an assistant that strips referrers produces sessions that look direct, and a bot that is blocked produces no log line, so each gap is recorded as unknown with its reason. And attribution is not causal lift: a conversion that followed an assistant referral was influenced, not caused, by it; lift needs a holdout or an experiment, which the plan states explicitly when it is and is not available.

Practically, the plan adds a channel grouping for known assistant referrers, a log-based report of AI user agents with their robots status, a monthly visibility sample read beside (never summed into) traffic, and a separate event design for any agent you operate. The dashboard labels each figure with its source and its denominator.

Attribution assigns credit for a conversion to the touchpoints before it. GA4 currently offers data-driven and last-click models; the older rule-based models were retired, as Google Analytics documentation notes. Google Ads, Meta and your CRM each apply their own model and window, so totals never match.

We treat attribution as a comparison tool, not a source of truth. No model observes journeys that happen off-site, across devices or without consent. For budget decisions we add BigQuery joins and geo holdout tests, and state the limits in every report.

We are a search and performance agency, so measurement is never an isolated project: the same key events feed our Google Ads bidding, social advertising reporting and SEO work, and our CRO and UX research starts from the behaviour data this setup produces. We work in Turkish, English, Arabic and Persian, which matters where consent banners, forms and conversion paths differ per locale, and we are conservative about what data can prove.