How NetEase Yanxuan Built an AI‑Powered Omnichannel Reach System

What Is Omnichannel?

In the second half of the internet era, traffic becomes increasingly fragmented. Yanxuan, a brand e‑commerce business, now has many traffic sources and sales channels: its own website, app, web and WAP, WeChat mini‑program, stores on Tmall, Taobao, JD.com, live‑stream platforms like Douyin and Kuaishou, as well as offline flagship and campus stores. Operating user flow across all these channels is a new challenge compared to platform‑centric e‑commerce.

Why Build an Omnichannel Reach System?

1. Omnichannel strategy is a natural choice for a brand e‑commerce. Every channel that can reach the target consumer is essential infrastructure.

2. Major e‑commerce platforms cannot share data with each other, so only a neutral third‑party can integrate data across channels.

3. Small third‑party providers lack data confidentiality, advanced algorithms, and deep e‑commerce experience, making them unsuitable for Yanxuan’s needs.

Therefore, building its own omnichannel reach capability is the optimal solution.

Yanxuan’s Omnichannel Reach Business Scenarios

Overall business goals:

Unified management of users across all channels.

Global ROI optimization for the entire user lifecycle.

Increase efficiency of user reach while improving user experience.

The system must, under given business goals and budget constraints, intelligently decide which users to target, what content to send, when and where to send it, and how much to spend, maximizing ROI.

Problems to Solve

Before the project started, four major issues existed:

1. Disconnected intervention measures

Different teams handled different reach scenarios (new‑user conversion, churn prevention, re‑activation, real‑time conversion, VIP acquisition) without awareness of each other, leading to fragmented interventions.

2. Missed optimal marketing timing

Traditional operations relied on manual tagging and batch interventions (often T+1), causing delays. Real‑time intent‑recognition models now enable immediate intervention when a user is about to churn.

3. Over‑reach disturbing users

Multiple channels often sent duplicate messages to the same user, especially during large promotions. A fatigue‑management service and DSP advertising reduce unnecessary touches.

4. Lack of personalized content

Tag‑based group targeting cannot achieve one‑to‑one personalization. AI‑driven matching of products, copy, and timing, plus automated creative generation, address this gap.

Solution Architecture

The solution is built on Yanxuan’s CRM and DSP systems and consists of four layers from bottom to top.

1. Data Product & Platform Layer

Provides a user data bus for real‑time status and entitlement, a fatigue‑management service, a DMP for user and product profiles, a marketing middle‑platform for campaigns and discounts, and a creative platform for assets.

2. Algorithm Service Layer

Includes AI scheduling for lifecycle and intervention decisions, real‑time budget allocation, personalized matching of product, copy, and timing, intelligent collage generation, intent‑recognition for conversion/churn probability, and dynamic entitlement generation.

3. Application System Layer

Hosts Yanxuan’s self‑built DSP ad‑delivery system (integrating NetEase, Tencent, Toutiao, and other media, plus MarketingAPI for Douyin and JD), automatic audio‑visual synthesis, and the main‑site CRM that connects the app, mini‑program, and offline stores.

4. Business Layer

Supports four core marketing activities: acquisition, activation, VIP card issuance, and sales conversion.

The architecture offers three advantages: precise, personalized targeting; full‑channel coverage with lifecycle‑aware scheduling; and real‑time budget optimization for maximum ROI.

Current Results and Future Work

Automation now covers 60% of tasks, with 34% being non‑rule‑based intelligent actions, and overall conversion has increased by 113%. Remaining work includes migrating more scenarios to intelligent reach, expanding the creative pool, and enhancing automatic content synthesis.

Future research will focus on cultivating internal user triggers (habit formation), lowering purchase friction, and leveraging scarcity, environment, anchoring, and coupon effects to boost motivation.