How Yanxuan Built an AI‑Powered Omnichannel Reach System to Maximize ROI

What is Omnichannel

In the later stage of the Internet, traffic is increasingly fragmented across many channels: a brand‑e‑commerce site’s own app, web, WAP, WeChat mini‑program, major marketplaces (Tmall, Taobao, JD.com), live‑stream platforms (Douyin, Kuaishou), and offline flagship or campus stores. Managing user flow across all of these channels requires a unified, data‑driven reach system.

Why an Omnichannel Reach System Is Needed

Omnichannel strategy is a natural evolution for a brand‑centric e‑commerce business; every channel that can reach the target consumer becomes essential infrastructure.

Major e‑commerce platforms are competitors and cannot share data, leaving only neutral third‑party providers capable of cross‑channel integration.

Small third‑party providers lack data confidentiality, advanced algorithmic research, and deep e‑commerce operational experience.

Core Business Scenarios

Unified management of users across all channels.

Global ROI optimization for the entire user lifecycle.

Increase reach efficiency while reducing waste and improving user experience.

Main Decision Flow

Identify target users based on business objectives.

Automatically calculate the lowest‑cost actions.

Locate the most timely touchpoints.

Select the most efficient timing.

Match personalized content (product, copy, creative) to each user.

Key Problems Addressed

1. Disconnected Interventions

Different teams handle scenarios such as new‑user conversion, churn prevention, re‑activation, high‑probability purchase, and super‑member acquisition. Without a shared data layer, interventions are unaware of each other, causing fragmented actions.

2. Missed Optimal Marketing Timing

Traditional batch labeling (often T+1 or weekly) delays interventions. Real‑time intent‑recognition models now detect imminent churn or conversion, enabling immediate, timely actions.

3. Over‑Touching Users

Multiple channels and scenarios can generate several messages per day, especially during large promotions, leading to user fatigue. A fatigue‑management service and a DSP advertising system shift many touches to less intrusive ad placements.

4. Lack of Personalized Content

Group‑based targeting cannot achieve true one‑to‑one personalization. Personalization algorithms, NLP‑generated copy, and intelligent image synthesis deliver content tailored to each user’s interests.

Solution Architecture

Layer 1: Data Products & Platform

Provides a user data bus for real‑time status and entitlement storage, a fatigue‑management service, a Data‑Management‑Platform (DMP) for user and product profiles, a marketing middle‑platform for campaigns and discounts, and a creative platform for asset management.

Layer 2: Algorithm Services

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

Layer 3: Application Systems

Integrates a self‑built DSP that aggregates media from NetEase, Tencent, Toutiao and other sources, plus MarketingAPI connections to Douyin and JD. Cloud‑music algorithms enable automatic audio‑video synthesis for Douyin ads, generating hundreds of daily plans from minimal manual input. Also includes a CRM serving the app, WeChat mini‑program, and offline stores.

Layer 4: Business Layer

Supports four core marketing activities: acquisition, activation, super‑member card opening, and sales conversion.

Advantages of the Solution

Precise, personalized targeting using integrated cross‑channel data.

Full‑channel coverage with lifecycle‑aware intelligent scheduling for global optimality.

Real‑time budget optimization that continuously adjusts spend based on channel, creative, and entitlement performance, maximizing ROI.

Current Results and Future Work

The system is still under construction. Automated tasks now account for 60% of operations, with 34% being non‑rule‑based intelligent tasks, and overall conversion has increased by 113%.

Remaining challenges include migrating additional scenarios to intelligent reach, expanding the creative pool, and enhancing automated material synthesis. Further research is needed to cultivate internal user triggers, lower purchase barriers, and stimulate motivation through scarcity, anchoring, and discount effects.