Churn Prediction Model Using Temporal Data Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional online concierge systems struggle to accurately attribute customer churn events to specific causal events, such as order fulfillment times or compensation amounts, which hinders the implementation of optimal remedial actions to maintain customer engagement.
Innovation Solution
The system trains a churn prediction model using historical customer interactions, segmenting data into different time intervals to decorrelate customer features from online concierge system features, allowing the model to output attribution scores for causal events and identify the most influential events contributing to churn.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional models predict customer engagement likelihood, then the system can identify at-risk customers, but the system cannot accurately attribute churn to specific causal events
Solution Approach 1:
The patent segments the model's output into multiple independent attribution scores, each corresponding to a specific causal event (e.g., fulfillment time, compensation amount). This segmentation allows the system to maintain overall churn prediction accuracy while simultaneously providing precise attribution to specific events, resolving the contradiction between prediction accuracy and information loss.
2Productivity
If the system uses conventional training data, then the model can predict churn probability, but the model cannot identify which events caused the churn
Solution Approach 1:
The patent adds a new dimension to the model output by introducing event-level attribution scores alongside the overall churn probability. This dimensional expansion allows the model to maintain its churn prediction functionality while simultaneously providing causal event identification, effectively resolving the information loss problem without sacrificing productivity.
3Reliability
If remedial actions are implemented without accurate event attribution, then the system can maintain general customer engagement, but cannot implement optimal targeted remediation
Solution Approach 1:
The patent implements a feedback mechanism where attribution scores inform the selection and customization of remedial actions. The system uses the event-level attribution information to provide targeted feedback to customers about specific issues (e.g., delayed fulfillment, low compensation) and applies针对性 remediation, thereby enhancing both engagement reliability and remediation ease.
Data Source
AI summary
An online system trains a churn prediction model to attribute a churn event to one or more causal events. The churn prediction model receives customer features and online system features as inputs. Various causal events that occur affect one or more online system features. To avoid biasing the churn prediction model using input features that are related to possible causal events, the online system determines customer features and online system features based on customer interactions occurring in different time intervals. The customer features are determined from interactions in a time interval that is earlier than a time interval from which interactions are used to determine online system features. Such time segmenting decorrelates the features input to the model from the events, reducing potential bias from the causal events on the churn prediction model.