Rebuilding Marketing Measurement for a Multi-Market DTC Retailer

Challenge

Ad platform numbers, GA4 and the finance report disagreed by a wide margin each month, so channel budget decisions were being argued rather than decided.

Business problem

Each month the ad platforms, GA4 and the finance report produced materially different revenue numbers. Budget meetings became arguments about methodology instead of decisions about spend, and two channels were being funded on self-reported credit nobody had ever tested.

Solution

A warehouse-backed measurement layer with server-side collection, identity resolution and a multi-touch model, validated against geo holdout tests.

Architecture

• Server-side collection layer handling consent mode and event validation
• Raw events, orders, costs and CRM records landed in BigQuery
• Identity resolution joining anonymous sessions to customers and orders
• dbt models with data tests producing a governed metrics layer
• Multi-touch attribution model built on the governed layer, not on raw platform data
• Looker Studio reporting plus geo holdout tests validating the model's conclusions

Technology stack

Implementation

• Rebuilt the tracking plan and event taxonomy from scratch
• Deployed server-side collection with consent mode handling
• Modelled orders, costs and touchpoints in dbt with data tests
• Ran geo holdout tests on the two largest channels

Results

• Reporting reconciled with finance within a low single-digit variance
• Two channels shown to be materially over-credited by platform reporting
• Tracking breakage now surfaced by alerts within a day

Business impact

Budget shifted toward channels with tested incremental effect, and the monthly reporting dispute ended.

Lessons learned

• Rebuilding the tracking plan first was unavoidable; modelling on top of an inconsistent taxonomy would have reproduced the original problem.
• Geo holdout testing changed the conversation entirely — modelled attribution alone would not have convinced the finance team.
• Data tests with alerting turned tracking breakage from a monthly surprise into a same-day fix.