Causal-driven attribution (CDA): Estimating channel influence without user-level data

TL;DR

The paper tackles the privacy-driven challenge of marketing attribution by introducing Causal-driven Attribution (CDA), a framework that operates on aggregated impression-level time series. CDA couples PCMCI-based temporal causal discovery with a Structural Causal Model for estimating direct and indirect channel effects, enabling causal attributions without user-level data. Through extensive synthetic-data experiments, CDA demonstrates accurate causal graph recovery (especially with the true graph) and meaningful, rank-preserving causal effect estimates even when the graph is learned rather than known, while highlighting robustness to structural uncertainty and sensitivity to deeper causal graphs. The approach offers a scalable, privacy-preserving alternative to path-based models and provides practical insights for budget allocation and cross-channel strategy in modern, privacy-conscious advertising environments.

Abstract

Attribution modelling lies at the heart of marketing effectiveness, yet most existing approaches depend on user-level path data, which are increasingly inaccessible due to privacy regulations and platform restrictions. This paper introduces a Causal-Driven Attribution (CDA) framework that infers channel influence using only aggregated impression-level data, avoiding any reliance on user identifiers or click-path tracking. CDA integrates temporal causal discovery (using PCMCI) with causal effect estimation via a Structural Causal Model to recover directional channel relationships and quantify their contributions to conversions. Using large-scale synthetic data designed to replicate real marketing dynamics, we show that CDA achieves an average relative RMSE of 9.50% when given the true causal graph, and 24.23% when using the predicted graph, demonstrating strong accuracy under correct structure and meaningful signal recovery even under structural uncertainty. CDA captures cross-channel interdependencies while providing interpretable, privacy-preserving attribution insights, offering a scalable and future-proof alternative to traditional path-based models.

Figures (8)

• Figure 1: Example of a DAG's structure, including nodes: Affiliates, Facebook, TikTok, Google Ads, YouTube, and conversion. Edges represent directed causal influences: green arrows indicate positive values (putative amplification/halo), and red arrows indicate negative values (putative substitution/crowd-out). Nodes are grouped into causal layers as follows: $S^3$: Affiliates, $S^2$: Facebook, TikTok, Google Ads, $S^1$: YouTube, $S^0$: conversion.
• Figure 2: Example of a causal DAG showing the structural relationships among marketing channels. Some channels act as upstream drivers that influence conversions indirectly through intermediate touchpoints, while others have a direct impact on the outcome. All effects flow in a single direction with no circular feedback, providing a clear view of how influence moves through the system.
• Figure 3: Overview of the simulation framework.
• Figure 4: A random simulated data illustrating marketing channel impressions and conversions within 365 days. The top panel presents the impression trends for five marketing channels, including Google Ads, Affiliates, TikTok, Facebook, and YouTube. Each series follows a distinct stochastic pattern generated from a linear growth process with random noise and inter-channel influence.
• Figure 5: Example of a DAG output representing multi-channel interactions from randomly simulated data. Each node represents a marketing channel (Affiliates, Google Ads, TikTok, Facebook, YouTube) or the target outcome (Conversions). Directed arrows denote causal influence between nodes, where the direction of the arrow indicates the flow of information from cause to effect. Green arrows correspond to positive causal weights, indicating that an increase in the originating channel’s activity leads to a subsequent increase in the target channel or conversions.