LinkedIn has introduced an enhanced ad attribution process aimed at improving outcome tracking for marketing campaigns. This new approach utilizes a hybrid model that combines bottom-up and top-down methodologies to provide accurate data-driven attribution. This is essential for understanding customer journeys and optimizing marketing strategies to maximize return on investment (ROI).
Attribution Approaches
Attribution measurement generally falls into two categories: rule-based attribution (RBA) and data-driven attribution (DDA). RBA methods, such as first-touch and last-touch, assign conversion credit based on predetermined rules, which can lead to biased results, especially in B2B contexts. In contrast, DDA employs machine learning techniques like Multi-Touch Attribution (MTA) and Marketing Mix Modeling (MMM), offering a more comprehensive view of customer interactions throughout the buying journey.
LinkedIn's Unified System
LinkedIn has developed a unified attribution system that integrates both MTA and MMM methodologies. This system has been successfully implemented for internal marketing and will be extended to advertisers on the LinkedIn Marketing Solutions platform.
Modeling Framework
The attribution model uses attention-based modeling to predict conversion outcomes based on member and touchpoint data. The model is structured as a binary classification task, where the conversion probability is influenced by various factors, including member representation and marketing touchpoint sequences. Attention weights are derived from the model to indicate the contribution of each touchpoint to conversions.
Path Construction and Representation
The model processes marketing paths as time series data, capturing the sequence of touchpoints. It uses positional encodings to account for the order of touches and incorporates features related to both members and their companies. This allows for a nuanced understanding of interactions and their impact on conversions.
Calibration and Lift Measurement
To ensure consistency between MTA and MMM results, a post-modeling calibration step aligns attribution estimates with MMM outputs. Channel lift is measured by comparing conversion probabilities of original and counterfactual paths, providing insights into the effectiveness of marketing channels.
Application and Results
The new attribution platform has shown promising results, particularly for upper and mid-funnel campaigns, which were previously undervalued under last-click attribution models. Initial findings indicate a 150x increase in credit attributed to modeled attribution compared to last-click methods, contributing to an estimated 5% lift in marketing-driven revenue.
Future Directions
LinkedIn plans to further enhance its attribution capabilities by exploring methods to improve causal robustness and align simulated campaign effects with experimental data. The attribution library is also being made available for use across different teams within LinkedIn, fostering collaboration and cross-domain improvements.
In conclusion, LinkedIn's shift from rule-based to data-driven attribution has yielded valuable insights and improved the quality of marketing performance analysis, paving the way for more effective budget allocation and strategy optimization.
