Attribution Credit Assignment via Counterfactual Path Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current attribution models, such as last click attribution, fail to accurately assign credit to multiple media exposures contributing to a converting act, as they only credit the last exposure, neglecting the influence of earlier exposures.
Innovation Solution
A data-driven attribution model that calculates attribution credits based on counterfactual gains by analyzing visit-related data, determining conversion probabilities of path types, and assigning credits to each event in a path type based on its contribution to conversions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If last click attribution model is used, then simplicity of attribution model is maintained, but accuracy of conversion tracking deteriorates
Solution Approach 1:
The patent segments the conversion attribution process into multiple components by dividing attribution credit across multiple events in the conversion path. Instead of assigning all credit to a single last click event, the system segments credit allocation among all contributing events (e.g., display ad, search ad, referral) based on their individual contributions to the conversion, thereby improving measurement accuracy while maintaining model operational simplicity.
Solution Approach 2:
The patent changes the parameter of attribution credit allocation from a single-point assignment (100% to last click) to a distributed assignment model where credit is allocated across multiple events based on counterfactual analysis. This parameter change enables more accurate conversion tracking by reflecting the actual multi-touch nature of consumer decision-making processes.
2Device complexity
If credit is assigned only to last media exposure, then computational complexity is reduced, but attribution accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing counterfactual conversion probabilities for different event sequences before actual attribution is needed. The system pre-processes historical conversion data to establish baseline conversion rates for various path types, which are then reused during attribution calculations. This preliminary computation reduces real-time computational complexity while maintaining high attribution accuracy.
Solution Approach 2:
The patent introduces counterfactual conversion probability as an intermediary metric that mediates between raw event data and final attribution credits. Instead of directly calculating complex multi-event attribution relationships, the system uses counterfactual probabilities as an intermediate step to simplify computations. This intermediary layer enables accurate attribution by breaking down the complex calculation into manageable components that can be processed efficiently.
3Measurement precision
If all media exposures are analyzed for attribution, then attribution accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent segments the comprehensive set of all media exposures into distinct path types based on event sequences. By grouping events into standardized path types (e.g., display-search-referral, search-direct), the system reduces data processing complexity through categorization. Each path type can be analyzed using pre-established counterfactual models, enabling accurate attribution across all exposures without processing each individual event sequence from scratch.
Solution Approach 2:
The patent transforms the complex analysis of individual event sequences into a parameter-based approach where attribution is determined by path type characteristics rather than detailed event-by-event analysis. By changing the analysis parameter from granular event-level details to aggregated path type metrics, the system maintains high attribution accuracy while significantly reducing data processing complexity through parameter abstraction.
Data Source
AI summary
Systems and methods for creating rules for assigning attribution credit across events, includes, identifying, by a processor, conversions at a website. The processor identifies path types associated with the conversions. Each path type identifies events and a index position indicating an event's relative position. The processor identifies a subset of the identified path types to be rewritten according to a path rewriting policy. The processor then rewrites the identified subset of the identified path types as rewritten path types. The processor determines, for each of the rewritten path types and remaining identified path types associated with the identified conversions, attribution credits for each event included in the path type. The processor creates, for each of the rewritten path types and remaining identified path types associated with the identified conversions, a rule for assigning the determined attribution credit to each event of the path type for which the rule is created.