Intelligent Reach System: Modeling, Decision Making, and Optimization for E‑commerce

This article introduces the intelligent reach (触达) technology used in e‑commerce scenarios, focusing on how to decide the target audience, timing, channel, welfare, and creative content for each outreach task.

Background: Drawing from the concepts in *Growth Hacking*, the authors emphasize the importance of reach across the user lifecycle—acquisition, activation, retention, monetization, and re‑engagement. Reach actions include in‑site messages, pop‑ups, push notifications, SMS, and external ads.

Reach Classification: Reach is divided by location (in‑site vs. out‑of‑site) and by trigger (proactive vs. touch‑point). The four‑quadrant diagram illustrates these categories.

Key Elements of Proactive Reach: audience, timing, channel, welfare, and creative. An example from the Double‑11 promotion shows how each element is decided (e.g., selecting female users interested in apparel, sending SMS at 10 pm, and testing creative copy).

Challenges of Manual Decision‑Making: repetitive creatives, high configuration effort, priority conflicts, and lack of channel coordination.

Intelligent Reach System: The system consists of a planning module that generates reach schedules offline and a real‑time module for urgent tasks. The planning pipeline uses batch processing to decide audience, time, channel, welfare, and creative before execution.

Intelligent Decision Models: Three model families empower the system—user models (lifecycle, churn probability, purchase probability, membership prediction using XGBoost), content models (template + variable generation, personalized variables, title rewriting via NLP), and decision models (time selection, channel selection using Bayesian CTR estimation and integer linear programming, welfare allocation with linear programming under budget and ROI constraints).

Decision‑Making Details: Time: three stages—global active periods, individual active periods, and model‑based predictions. Experiments show that personalized time selection improves conversion. Channel: estimation of CTR per channel via Bayesian inference, followed by optimization to allocate budget between SMS and push. Welfare: two strategies—tiered discounts for high‑spending users and larger coupons for low‑activity users. Models address data sparsity, bias, and multi‑objective constraints.

Future Work: richer personalization of creatives, joint optimization of multiple elements, gain‑based welfare estimation, fatigue modeling, and cross‑channel coordination.

Conclusion: The intelligent reach system, built on relatively simple yet effective models, yields a 17.4% increase in paying users and a 5% lift in revenue compared with manual strategies.