Multi-Touch Attribution

Statistical method for distributing conversion credit across the marketing touchpoints a customer encountered before converting. Addresses the core measurement question in performance marketing: which channels, campaigns, or interactions drove a given outcome.

Attribution models. Rules-based models are computationally cheap and interpretable:

• First-touch: 100% credit to the first interaction.
• Last-touch: 100% credit to the final interaction before conversion.
• Linear: Equal credit split across all touchpoints in the path.
• Time-decay: Exponentially increasing credit to touchpoints closer to the conversion event.
• Position-based (U-shaped): Fixed allocation to first and last touch (typically 40/40), remainder distributed across mid-path touches.

Algorithmic models use transition probability or game-theoretic frameworks:

• Markov chain: Models the customer journey as a state-transition graph; channel credit is proportional to the reduction in conversion probability when a given channel is removed from the graph.
• Shapley value: Game-theoretic method; assigns each channel the average marginal contribution across all possible orderings of touchpoints in the conversion path.

CDW implementation patterns. In composable CDP stacks, MTA is typically built as dbt models over event-stream tables (web events, email clicks, ad impressions, purchase events) joined on a persistent customer ID. The Tasman Analytics open-source dbt package (tasmananalytics/tasman_dbt_mta) provides a reference implementation using both Markov chain and Shapley value algorithms over a standard events-and-conversions schema.

For teams using Snowflake ML, Snowflake Model Explainability (via the Snowflake Model Registry) provides Shapley-value attribution for trained conversion-prediction models — applicable when the attribution question is framed as explaining which features (channels, events) drove a model's conversion-probability score, rather than as a raw touchpoint-credit pipeline.

CDP integration. MTA pipelines consume event data routed through the CDP collection layer (Segment Connections, Tealium EventStream, AEP Web SDK). Attribution outputs — channel-level credit scores and conversion path summaries — are typically published to BI and reporting destinations (Snowflake → dbt → Tableau, Looker, or Cortex Analyst) rather than fed back into activation pipelines. See capability.streaming-ingestion for the upstream event capture layer.

Practitioner context (2026): Unified Marketing Measurement. In 2026, standalone MTA is a minority pattern among mature marketing measurement organizations. The dominant architecture is Unified Marketing Measurement (UMM): MTA and Media Mix Modeling (MMM) run simultaneously and their outputs are reconciled. MTA provides granular, privacy-safe, consent-grounded attribution for digital touchpoints at the individual-user level. MMM provides aggregate modeling of non-digital channels (linear TV, out-of-home, radio) and long-term trend signals that MTA cannot capture from touchpoint sequences. CDP event pipelines feed the MTA model; aggregate media-spend data feeds the MMM model. The integration layer — where individual-level MTA credit scores are reconciled with channel-level MMM coefficients — is typically implemented in Snowflake, dbt, or a purpose-built MMM platform (open-source options include Google Meridian and Meta Robyn). Evaluators building a composable CDP measurement stack should plan for MTA as one component of a larger UMM system rather than a standalone attribution tool.