Abstract:Modern e-commerce services frequently target customers with incentives or interventions to engage them in their products such as games, shopping, video streaming, etc. This customer engagement increases acquisition of more customers and retention of existing ones, leading to more business for the company while improving customer experience. Often, customers are either randomly targeted or targeted based on the propensity of desirable behavior. However, such policies can be suboptimal as they do not target the set of customers who would benefit the most from the intervention and they may also not take account of any constraints. In this paper, we propose a policy framework based on uplift modeling and constrained optimization that identifies customers to target for a use-case specific intervention so as to maximize the value to the business, while taking account of any given constraints. We demonstrate improvement over state-of-the-art targeting approaches using two large-scale experimental studies and a production implementation.

What problem does this paper attempt to address?

The paper attempts to address the problem of how to more effectively intervene with customers in e-commerce to enhance business value while considering various business constraints. Specifically, existing customer intervention strategies are often based on customer propensity scores (e.g., risk of customer churn), but these strategies may not be optimal because they do not consider which customers would benefit the most from the intervention, nor do they fully account for business constraints (such as budget limitations). Therefore, the authors propose a framework based on uplift modeling and constrained optimization, aiming to identify the customer groups most likely to respond positively to specific interventions, thereby maximizing business returns. ### Main Contributions of the Paper: 1. **Proposed a structured two-step approach**: This method first estimates the response effect of each customer to the intervention (i.e., uplift value) and then finds the optimal target customer group under business constraints. 2. **Demonstrated the application of this method in three different business scenarios**: Each scenario has unique business objectives and operational constraints. 3. **Proved the superiority of this method over commonly used propensity score threshold-based targeting methods** through offline policy evaluation techniques in three key business applications. 4. **Validated the effectiveness of the proposed policy through large-scale online A/B testing**. ### Specific Application Scenarios: 1. **Increasing customer retention rate**: By identifying customers most likely to respond positively to interventions, improve customer retention rates. 2. **Maximizing campaign revenue**: By optimizing the way customers participate in shopping events and promotions, increase total revenue while controlling costs. 3. **Optimizing reward allocation in shopping events**: By adjusting the purchase thresholds for different customers, maximize total revenue while maintaining customer experience. ### Method Framework: 1. **Uplift Model**: Used to estimate the response effect of each customer to the intervention. 2. **Constrained Optimization**: Finds the optimal target customer group under business constraints. ### Evaluation Metrics: - **Uplift Model Evaluation**: Use the area under the cumulative uplift curve (AUC) to evaluate model performance. - **Offline Policy Evaluation**: Use metrics such as Inverse Probability Weighting (IPS) and Self-Normalized Inverse Probability Weighting (SNIPS) to evaluate the effectiveness of different policies. Through these methods and evaluation metrics, the authors demonstrate the effectiveness and superiority of the proposed framework in practical applications.

Segment Discovery: Enhancing E-commerce Targeting

A Privacy-Aware Framework for Online Advertisement Targeting

Delivery Optimized Discovery in Behavioral User Segmentation under Budget Constraint

Personalized Mobile Targeting with User Engagement Stages: Combining a Structural Hidden Markov Model and Field Experiment

Optimizing Scalable Targeted Marketing Policies with Constraints

Enhancing geotargeting with temporal targeting, behavioral targeting and promotion for comprehensive contextual targeting.

Entire Chain Uplift Modeling with Context-Enhanced Learning for Intelligent Marketing

A Model-based Projection Technique for Segmenting Customers

Modeling User Engagement in Mobile Content Consumption with Tapstream Data.

Targeted Promotions on an E-Book Platform: Crowding Out, Heterogeneity, and Opportunity Costs

Dynamically Tie the Right Offer to the Right Customer in Telecommunications Industry

Targeted Promotions and Cross-Category Spillover Effects

Don’t Miss the Potential Customers! Retrieving Similar Ads to Improve User Targeting

When and How to Leverage E-commerce Cart Targeting: the Relative and Moderated Effects of Scarcity and Price Incentives with a Two-Stage Field Experiment and Causal Forest Optimization.

Enhancing Uplift Modeling in Multi-Treatment Marketing Campaigns: Leveraging Score Ranking and Calibration Techniques

UNICON: A unified framework for behavior-based consumer segmentation in e-commerce

A review on customer segmentation methods for personalized customer targeting in e-commerce use cases

Deep Reinforcement Learning for Sequential Targeting

Coarse Personalization

Multidimensional Targeting and Consumer Response

Targeting for long-term outcomes