Configure a Multi-Touch Attribution Model
Define and run an algorithmic model that assigns fractional revenue or conversion credit across all touchpoints in a customer journey.
Multi-touch attribution distributes credit for a conversion across all the marketing touchpoints a customer encountered before completing the goal action. Unlike last-touch or first-touch heuristics, an algorithmic model uses the full path of interactions to learn which combinations of channels and sequence patterns are most causally predictive of conversion. The output is a per-touchpoint fractional credit value that can be aggregated across campaigns to calculate true incremental return on ad spend.
Configuring the model requires the practitioner to specify: which event type represents a conversion (purchase, form submit, subscription), optionally a monetary value field for weighted revenue attribution, the look-back window (how far back in the journey to consider), and the channel fields that identify each touchpoint category. Run frequency should match the pace of campaign changes — monthly for stable evergreen programs, weekly for high-rotation performance marketing.
Parallel viability note: Dedicated multi-touch attribution tools (e.g., Rockerbox, Northbeam, Triple Whale) are vendor-neutral alternatives that operate on event-level data feeds and return score datasets compatible with any analytics warehouse. Snowflake and BigQuery can host custom Shapley-value implementations. AEP's Attribution AI is tightly integrated with the XDM event schema and writes directly to the Platform dataset layer, which reduces pipeline complexity when the primary profile store is AEP but adds a dependency on the Intelligent Services licensing tier.
Side-by-side implementations
Parallel implementation not yet available.
Snowflake supports multi-touch attribution via two complementary mechanisms. For algorithmic (data-driven) attribution, Snowflake ML Model Explainability — available in the ML Model Registry — computes Shapley values for trained ML models; these Shapley-value feature attributions can be applied to channel-level credit allocation by training a conversion prediction model on session-ordered touchpoint features and extracting per-channel Shapley contributions. For rule-based MTA (first-touch, last-touch, linear, U-shaped, time-decay), the tasman-dbt-mta open-source dbt package (Tasman Analytics; v1.1.2, April 2026; GPL-3.0) implements each attribution model as a set of dbt models operating on an event table with session_id, channel, and conversion_flag columns. The package supports Snowflake and BigQuery natively. The attribution output is materialized as a dbt-scheduled Snowflake table and feeds a Snowflake-hosted BI layer (Tableau, Looker, Mode) or exports via Hightouch to a downstream ad platform conversion API. If tradeoff.data-egress applies (Snowflake egress for external ad platform export vs. AEP bundled pricing), add tradeoffs: [tradeoff.data-egress] to the node.
Capability: Audience Segmentation
Parallel implementation not yet available.
Task-level sources
- technical-training/module5/index.md
- technical-training/module5/ex1.md
How is this implementation?
Sign-in-gated. Tomorrow morning's curriculum-ingestor consumes your feedback: "Inaccurate" queues the task for re-review, "needs update" queues it for a refresh, and "one vendor panel is wrong" re-drafts just that panel.