Synthetic Control Group Generation for Ad Program Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing the effectiveness of computer-driven communication programs, such as difference in differences (DID) techniques, require significant additional computer resources to identify and track independent control and experimental groups, leading to inefficiencies in data collection, analysis, and program reconfiguration.
Innovation Solution
A system, method, and computer program product for generating a synthetic control group, which involves receiving transaction account data, generating a synthetic control group by sampling from the transaction accounts, determining propensity scores using machine learning, assigning entropy balancing weights, and altering operational parameters of a computer-implemented advertisement program based on the synthetic control group.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If DID techniques are used to analyze communication program effectiveness by comparing exposed and non-exposed groups, then measurement precision is improved, but device complexity and loss of time increase due to requiring identification and tracking of two independent groups across multiple time periods
Solution Approach 1:
The patent creates a synthetic control group by sampling and weighting transaction accounts to replicate the characteristics of a true control group, rather than requiring an actual independent control group. This copying approach maintains measurement precision while eliminating the complexity of tracking separate groups.
Solution Approach 2:
The patent transforms the control group identification problem by changing from tracking actual control groups to generating synthetic control groups through propensity score sampling and entropy balancing weights. This parameter transformation resolves the contradiction by maintaining analytical rigor while reducing operational complexity.
2Measurement precision
If DID techniques require identification and tracking of two independent groups from multiple time periods, then measurement precision is improved, but loss of time increases due to additional data observation requirements
Solution Approach 1:
The patent performs preliminary sampling and weighting of transaction accounts to create the synthetic control group before the actual analysis is needed. This preliminary action eliminates the need for time-consuming tracking of independent control groups across multiple periods, as the synthetic group is pre-configured to represent the control population.
Solution Approach 2:
By creating a synthetic copy of the control group through sampling and weighting, the patent eliminates the need for actual long-term tracking of separate control groups. The copied synthetic group maintains the statistical properties needed for precise measurement without the time cost of actual longitudinal tracking.
3Measurement precision
If additional computer resources are allocated to identify and track independent control groups, then measurement precision is improved, but productivity decreases due to reduced efficiency in data collection and analysis
Solution Approach 1:
The patent replaces resource-intensive tracking of actual control groups with a computationally efficient synthetic control group generated through sampling and weighting algorithms. This copying approach maintains measurement precision while dramatically improving productivity by eliminating the need for continuous resource allocation to group tracking.
Solution Approach 2:
The patent substitutes the mechanical process of identifying and tracking independent control groups with a computational algorithm that generates synthetic control groups through propensity score sampling and entropy balancing. This substitution replaces resource-heavy operational procedures with efficient computational methods, improving productivity while maintaining precision.
Data Source
AI summary
Described are a system, method, and computer program product for generating a synthetic control group. The method includes receiving transaction data associated a first set of transaction accounts in a first time period and generating a synthetic control group including a subset of transaction accounts. The method also includes determining, for each transaction account of the synthetic control group, a propensity score representative of a likelihood of said transaction account being associated with a test group, and a predictive spending score for a second time period. The method further includes balancing the synthetic control group and altering at least one operational parameter of a computer-implemented advertisement program to be executed in a second time period based on the transaction data and the synthetic control group. The method further includes executing the computer-implemented advertisement program in the second time period based on the at least one operational parameter that was altered.