Conversion Timing Prediction Using Gamma Mixture Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current attribution schemes in networked advertising struggle to accurately predict conversion timing, making it difficult for advertisers and online vendors to take timely and appropriate actions, such as setting bid prices or providing customized content.
Innovation Solution
A conversion timing system that models conversion timing by fitting a gamma mixture model to the distribution of conversion timespans, allowing for the prediction of conversion likelihood based on elapsed time since a qualified entry event, enabling timely and customized responses to potential customers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional attribution schemes are used to track conversion timing, then the system is simple to implement, but the prediction accuracy of conversion timing is poor
Solution Approach 1:
The patent transforms the conversion timing prediction problem from a simple attribution task to a probabilistic modeling task by changing the parameter representation. Instead of using binary attribution (converted/not converted), the system models conversion timing as a continuous probability distribution using gamma distribution parameters (shape and scale), enabling precise timing predictions while maintaining computational efficiency through parameter-based representations rather than complex simulations.
Solution Approach 2:
The patent introduces conversion timing models as intermediary components between the raw conversion data and the advertising decisions. These models act as mediators that process elapsed time information and convert it into predicted conversion probabilities, bridging the gap between simple timing tracking and complex conversion optimization without requiring direct implementation of sophisticated attribution algorithms at every decision point.
2Speed
If conversion timing prediction is implemented to enable timely actions, then the responsiveness to customer opportunities is improved, but the computational processing time increases
Solution Approach 1:
The patent performs preliminary actions by pre-calculating and storing conversion timing models during periods when computational resources are abundant. The gamma distribution models are fitted to historical conversion data in advance, so that when real-time advertising decisions are needed, the system only needs to query pre-computed models rather than performing complex calculations, thus achieving fast response times without sacrificing prediction accuracy.
Solution Approach 2:
The patent replaces complex mechanical attribution tracking systems with a statistical modeling approach. Instead of using intricate rule-based attribution logic that requires extensive processing of multiple touchpoints, the system substitutes a probabilistic gamma distribution model that can be evaluated efficiently with simple mathematical operations, reducing computational overhead while maintaining predictive power.
3Loss of information
If detailed conversion timing data is collected and analyzed, then the understanding of customer behavior is improved, but the data processing complexity increases
Solution Approach 1:
The patent extracts only the essential information needed for conversion timing prediction from complex customer behavior data. Instead of processing and storing all raw interaction data, the system extracts key features such as elapsed time since entry event and fits these to gamma distribution parameters. This extraction approach retains the critical temporal patterns in customer behavior while discarding redundant information, simplifying the data processing system.
Solution Approach 2:
The patent transforms detailed customer behavior data into simplified parametric representations using gamma distribution parameters (shape and scale). By changing the data representation from raw event sequences to distribution parameters, the system maintains the essential behavioral patterns while reducing data complexity, enabling efficient processing and analysis without losing critical conversion timing information.
Data Source
AI summary
A conversion timing model is model is configured to predict a likelihood of conversion based on an entity's elapsed time since a qualified entry event and based on a funnel state. The conversion timing model is constructed based on a distribution of the conversion timespans of converters. A notification of an opportunity to expose a candidate entity to networked content is received. A time-based likelihood of conversion for the candidate entity is determined by applying the conversion timing model to the elapsed time. A response to the notification based on the likelihood of conversion for the candidate entity is prepared based on the time-based likelihood of conversion and based on the funnel state. Timely responses may include the selection of customized content or bid values.