Cross-Session Attribution Requires a Graph, Not a Funnel

Rule-based attribution models — first touch, last touch, linear, even position-based — assume user journeys are sequential. They're not. Users bounce between devices, revisit from different channels, and convert weeks after initial exposure. Journey data shows that the clean funnel model is a useful fiction, not a reflection of reality.

Build an attribution graph in BigQuery using user_pseudo_id (which identifies a browser or app instance, not a person) or, better, a deterministic cross-device identifier such as a logged-in user_id. Model channels as nodes and the transitions between touchpoints as edges, with decay-weighted influence based on recency and interaction depth (scroll depth > pageview > impression).

Use BigQuery ML's logistic regression to derive data-driven weights from your own conversion data rather than trusting GA4's black-box DDA model. The resulting attribution is specific to your business, explainable to stakeholders, and auditable.

The implementation starts with a touchpoint table: each row is a user-channel-timestamp triple. From this, you build a transition matrix showing the probability of moving from one channel to another. The removal effect — what happens to conversion probability when you remove a channel — is your measure of each channel's contribution.

Present the channel graph to stakeholders. It's the most honest picture of how your marketing channels interact: which channels feed each other, which ones end journeys without a conversion, and where the handoffs that lead to conversion happen.