Measurement architecture & QA

We document the event schema your organisation intends to defend in audits and board packs: naming conventions, required parameters, default values, and the difference between marketing identifiers and product telemetry that must never mix. Each event carries an owner, a deprecation policy, and a consumer list so renames do not orphan half the warehouse.

Identity resolution rules are spelled out where they hurt most: how anonymous traffic becomes known, how merges are deduplicated, what happens when a user deletes an account mid-funnel, and which keys are considered authoritative in a conflict. That section is written so your data platform team can reconcile it with dbt tests or warehouse constraints without translating marketing language first.

Consent boundaries are mapped to tags, regions, and surface types—not as a policy PDF alone, but as rules engineers can implement in the tag manager and server-side collectors you actually run. We include a matrix of vendor defaults versus your legal posture so you can see where a platform’s “recommended setup” would over-collect relative to your register of processing activities.

Instrumentation & environments

Staging, preview, and production each receive explicit tagging policies: which events fire where, which destinations receive synthetic data, and how to prevent test traffic from polluting paid attribution or executive dashboards. When you ship dark launches or percentage rollouts, we document which events remain valid and which must be suppressed until the feature flag reaches stable state.

• Event dictionary with types, required properties, PII class, and example payloads.
• Consent mode / CMP wiring notes tied to vendor IDs you operate today.
• Backfill and replay procedures when a bug zeroes a day of collection.
• Data retention by stream, aligned to DPA schedules you already publish.

QA before every release

Regression scripts and synthetic journeys exercise critical funnels after each deploy, with failure thresholds that block a go-live instead of hiding under a Monday morning spike investigation. We align those checks with how your CI/CD pipeline already gates application releases so marketing instrumentation is not a special snowflake process.

We define “golden journeys” per product line—signup, upgrade, cancellation, refund—and attach expected event sequences with tolerances for timing skew when third-party pixels load slowly on mobile networks. Failures emit actionable artefacts: HAR snippets, tag load order, and the diff of your container publish so engineers do not chase ghosts.

Dashboards are treated as products: named owners, refresh SLAs, known limitations, and footnotes for sampling bias so PMMs and analysts argue from the same export hash instead of from screenshots taken three hours apart. Where a metric cannot be trusted yet, we label it “directional” and specify the engineering milestone that upgrades it to “contractual.”

Governance & handover

We leave you with a living measurement handbook—diff-friendly, versioned next to your repo or wiki—so new hires inherit the contract between marketing, legal, and data without a private Slack archaeology project. Quarterly review hooks list which events drifted most, which dashboards nobody opened, and which vendor invoices no longer match active integrations.

When you want to extend coverage or tighten SLAs, the same document anchors scope for the next engagement instead of restarting discovery from zero. Optional retainers cover measurement office hours: sign-off on new campaigns, review of vendor release notes that touch your stack, and incident support when a platform silently changes attribution defaults.