Attribution & ML

Multi-Touch Attribution Platform

Built ML-powered attribution system combining multi-touch modeling, incrementality testing, and MMM. Achieved 91% attribution accuracy and 18% CAC reduction.

AnalyticsMachine LearningConsultingData Engineering

Multi-Touch Attribution Platform - Featured visualization

Key Results

91% accuracy

18% CAC reduction

23% efficiency gain

Technology Stack

BigQuery ML ML Analytics dbt Data Transform OWOX BI Marketing Analytics Custom MMM Attribution Model Looker BI Dashboard

Step 1

The Challenge

A direct-to-consumer brand spending $50M+ annually on marketing made every budget decision based on last-click attribution—the default in their analytics platform. Deep analysis revealed the devastating impact: awareness channels (podcasts, influencers, brand video) showed terrible ROAS and were being systematically defunded. Meanwhile, retargeting and branded search—which simply captured demand created elsewhere—received ever-increasing budgets. They were optimizing for the bottom of a funnel while starving the top. But switching attribution models isn't simple—different models tell different stories, and leadership needed confidence in whatever replaced last-click.

Step 2

Strategic Approach

We designed a comprehensive attribution intelligence system combining multiple methodologies: multi-touch attribution for digital touchpoint credit, incrementality testing for causal measurement, and marketing mix modeling for understanding brand and offline channels. No single model tells the complete truth—but triangulating across methodologies reveals what's actually driving business outcomes. Our approach would give leadership not just new numbers, but confidence in those numbers through validation across methods.

Multi-Touch Attribution Platform - System architecture diagram

System architecture and workflow visualization

Step 3

Implementation Details

BigQuery ML powers custom attribution models trained on the brand's specific customer journey data. We implemented data-driven attribution that learns actual path-to-conversion patterns rather than applying arbitrary rules. The model weights touchpoints based on their actual influence on conversion probability.

dbt transformation pipelines prepare data from 8 sources—ad platforms, CRM, e-commerce, customer data platform—creating the unified customer journey dataset required for accurate attribution. Identity resolution connects anonymous website visits to eventual purchasers.

OWOX BI provides the multi-touch attribution layer, handling the complexity of cross-device journeys and long consideration windows. We configured custom channel groupings reflecting how the brand actually thinks about their marketing mix.

For brand and offline channels, we implemented Marketing Mix Modeling—statistical analysis of aggregate performance across channels, accounting for seasonality, competitive activity, and economic factors. Incrementality testing through geo-experiments validated the models: holding out markets from campaigns to measure true lift.

Looker dashboards present attribution data in business terms—not just which channels get credit, but what the optimal budget allocation should be and expected impact of reallocation.

Multi-Touch Attribution Platform - Implementation details

Technical implementation and integration details

Step 4

Measurable Results

Eight weeks of implementation and validation delivered transformative insights:

91% attribution accuracy validated through holdout testing, compared to 64% with last-click
23% improvement in marketing efficiency after reallocation based on true attribution
40% upper-funnel underinvestment discovered and corrected
18% CAC reduction through optimized channel mix
8 data sources unified into coherent customer journey view

The brand now makes $50M+ in annual marketing decisions with confidence in the underlying data.

Multi-Touch Attribution Platform - Results dashboard

Performance metrics and results visualization

Insights

Key Takeaways

Attribution is a measurement problem, not just a technology problem—multiple methodologies are required to triangulate truth. Last-click attribution systematically undervalues awareness channels, leading to funnel starvation over time. Incrementality testing through holdout experiments provides the ground truth needed to validate attribution models. Organizations that invest in proper attribution infrastructure make dramatically better marketing decisions than those relying on platform defaults.

Back to Portfolio

About My Approach View All Services