Building the Measurement Stack

A multi-year project. Five layers, the people who build them, and the governance that keeps it coherent.

Why building the stack is multi-year

The modern measurement stack isn't a tool you buy. It's an architecture: data collection on the customer-facing edges, a warehouse-centered data platform, analytics tools that consume warehouse data, activation tools that push insights back to operational systems, and a measurement layer that uses everything. Most organizations need 2–4 years to mature.

The five layers

1. Collection. Browser, server, mobile, in-store, CRM, ad platform events flow in.
2. Warehouse and modeling. Raw events land; transformations build derived tables.
3. Analytics and BI. Tools query the warehouse to produce reports and dashboards.
4. Activation. Audiences and insights pushed back to operational platforms.
5. Measurement. MMM, MTA, incrementality, brand tracking on top of everything else.

Data collection layer

• Web: Browser-side via GTM (gtag/GA4 base tag, pixel scripts). Server-side via GTM Server or custom backend.
• Mobile: Firebase SDK for app events; Branch / AppsFlyer / Adjust for attribution.
• Server: Direct API events from your backend (purchases, lead-to-customer, subscription events).
• CRM: Salesforce, HubSpot connectors to warehouse (Fivetran, Stitch).
• Ad platforms: Spend, impressions, clicks, platform-reported conversions (Fivetran, Funnel, Supermetrics, Adverity).
• Product analytics: Segment, RudderStack, Heap, Amplitude as event capture layers.
• Email and lifecycle: Klaviyo, Iterable, Customer.io exports to warehouse.
• In-store / offline: POS exports; loyalty data feeds.

Warehouse and modeling layer

• Warehouse: Snowflake, BigQuery, Databricks, Redshift. The center of gravity for all analytics.
• Transformation: dbt (or SQLMesh, Dagster) for version-controlled, tested SQL transformations.
• Modeling pattern: Source tables (raw) → staging tables (light cleanup) → intermediate tables (business logic) → marts (analyst-friendly tables).
• Identity resolution: User-stitching across email, device, session, account; built as warehouse models.
• Data quality: Tests on every dbt model; alerting on anomalies; documentation in dbt docs.

Analytics and BI layer

• BI tools: Looker, Tableau, Power BI, Mode, Hex, Metabase. Pick one canonical tool to prevent metric proliferation.
• Self-service analytics: Pre-built dashboards for stakeholder groups; ad-hoc query access for analysts.
• Semantic layer: dbt metrics, LookML, Cube.dev. Defines metrics once, consumed by all BI.
• Embedded analytics: Some BI tools embed dashboards into internal tools or customer-facing products.

Activation layer

• Reverse ETL: Hightouch, Census push warehouse data back to Meta, Google, Klaviyo, Salesforce, etc.
• CDP: Segment, Hightouch, mParticle for identity resolution and audience activation across channels.
• Server-side tagging: GTM Server-Side as alternative to CDP for simpler stacks.
• Use cases: Audience syncs to ad platforms, abandoned-cart triggers, churn-risk targeting, CRM enrichment.

Measurement layer

• MMM platforms: Recast, Haus, Northbeam, or in-house Robyn/LightweightMMM.
• MTA platforms: Platform-native (GA4 DDA), or vendor (Rockerbox, Triple Whale), or DIY in warehouse.
• Incrementality: Vendor (Measured, Haus, Recast) or DIY geo holdouts; platform-native lift studies.
• Brand measurement: Brand tracker surveys, share-of-voice, share-of-search.
• Custom analyses: Cohort retention, LTV modeling, churn modeling, optimization simulations.

Governance

• Data governance: Data dictionary, ownership, freshness SLAs, deprecation policy.
• Metric governance: Single source of truth for KPIs; documented definitions; semantic layer enforcement.
• Privacy governance: PII handling, retention, deletion compliance.
• Access governance: Role-based access; PII restricted to authorized roles.
• Change governance: Schema changes, metric changes, model changes communicated to consumers.

Advanced playbook

• Stack maturity assessment annually. Score each layer on coverage, quality, tooling. Identify gaps and plan next year's investment.
• Data contracts. Formal agreements between data producers and consumers; enforced in CI. Prevents silent schema breakage.
• Semantic layer as discipline. Don't let metrics live in 12 BI tools with 12 definitions. One semantic layer.
• dbt models as code review. Every transformation in version control with PR review. The same standards as application code.
• Identity resolution maturity. Move from cookie-only to email-stitched to account-stitched to fully unified customer profiles over time.
• Cost monitoring. Warehouse and tool costs scale with data volume. Quarterly cost reviews; identify wasteful queries and tools.
• Stakeholder enablement. Self-service BI works when users are trained. Invest in enablement; don't just provide tools.
• Vendor consolidation when sensible. Specialized tools accumulate; periodic consolidation reduces fatigue.
• Build the people, not just the tools. Analytics engineering, data engineering, marketing analytics are distinct skill sets. Hire intentionally.
• Documentation as deliverable. dbt docs, BI documentation, runbooks. If it's not documented, it's tribal knowledge that walks out the door.

Common mistakes

• Buying tools without architecture; tools accumulate without integration.
• No data warehouse; analytics scattered across vendor platforms with no consolidation.
• No transformation layer; raw data queried directly, breaks at scale.
• Multiple BI tools with diverging metric definitions; reports don't reconcile.
• No identity resolution; users counted multiple times across systems.
• Reverse ETL without governance; audience syncs become chaotic.
• Measurement layer (MMM, MTA, incrementality) bolted on without warehouse foundation.
• No data quality testing; bad data flows downstream silently.
• Cost runaway; queries optimized for convenience over efficiency.
• No documentation; new analysts can't onboard.
• Tool decisions without business case; vendor sprawl.
• Treating the stack as a project with an end date; it's ongoing infrastructure.

Operating checklist

• Data warehouse with all major sources flowing in
• Transformation layer (dbt or equivalent) with version control and tests
• Identity resolution model in the warehouse
• Canonical BI tool with semantic layer
• Reverse ETL for activation
• Measurement layer: MMM, MTA, incrementality, brand
• Privacy and access governance
• Data quality monitoring with alerting
• Cost monitoring and quarterly optimization
• Documentation: dbt docs, BI catalogs, runbooks
• Stack maturity assessment annually
• People: analytics engineers, data engineers, marketing analysts as distinct roles

---

Part of the Attribution & Measurement series.