How Low-Cost AI Powers Full-Scale Store Digitalization

Store Digitalization Overview

On November 25, Li Yongxiang, Technical Director of Suning Technology Group's R&D Center, presented the "Low‑Cost Empowerment of Store Digitalization" course, emphasizing that Suning's fully digital visual unmanned store is the starting point for large‑scale retail digital transformation, not the final AI‑based store model.

Key Topics Covered

Content of store digitalization

Unmanned store technology empowering digitalization

Store digitalization solutions

Store Digitalization Content

Current offline retail faces three main challenges: shifting consumer demands (post‑90s, post‑95s, post‑00s customers prioritize experience and quality over price), increasing product management complexity (tens of thousands of SKUs, long‑tail items, costly manual inventory), and rapid expansion of store formats (over ten business types and tens of thousands of locations). These challenges require new technological approaches.

Technology‑Driven Approach

Suning leverages big data, cloud computing, and AI across six product dimensions: precise recommendation, interactive marketing, reverse customization, unmanned inventory, digital site selection, and scene monitoring.

Product and Technology Matrix

The unmanned store R&D focuses on four technical pillars: image algorithms, recommendation algorithms, hardware capabilities, and media terminals.

Image Algorithms

Developed capabilities include facial recognition, pedestrian detection, generic object detection, and product recognition. Challenges addressed are unsupervised learning, incremental learning, crowded‑occlusion tracking, human‑product interaction analysis, and product registration mechanisms. For example, precise recommendation uses facial ID, pedestrian trajectories, and interaction analysis to link users with products.

Recommendation Algorithms

These algorithms build comprehensive user profiles and real‑time offline‑to‑online recommendation systems, extending internet e‑commerce recommendation techniques to physical stores by closing the loop between visual ID, behavior trajectories, and online actions such as browsing and adding to cart.

Hardware Capabilities

Key components are multi‑sensor, multi‑camera data acquisition and edge acceleration boxes. Edge devices enable low‑cost deployment of AI models on‑site, reducing reliance on high‑power GPU servers.

Media Terminals

Large screens, SMS, mobile apps, and WeChat are used for A/B testing and delivering personalized marketing, completing the recommendation‑marketing loop.

Unmanned Store Technology Empowering Store Digitalization

The "one‑person‑one‑record" concept unifies online and offline user IDs, sharing the same profile across Suning's e‑commerce platforms and physical stores. Data from facial capture cameras, 3D depth cameras, and standard security cameras provide comprehensive foot‑traffic and trajectory information, while multi‑pose registration ensures accurate ID matching.

Automatic Product Modeling

Product recognition faces challenges due to massive SKU counts and frequent changes. Suning’s automatic modeling algorithm combines sensor data, visual images, and shelf planograms, using weight sensors as semi‑supervised signals to iteratively update models, reducing manual labeling and achieving high accuracy in real‑time inventory.

With product detection, three business capabilities emerge: shelf compliance checking, product conversion heatmaps, and real‑time or batch inventory counting for both unmanned and conventional stores.

Dual‑Line Recommendation Fusion

By capturing offline behaviors such as shelf visits and product handling, Suning maps these signals to online recommendation models, enabling vector‑embedding based similarity matching for both user and product profiles. Recommendations are delivered via in‑store large screens and app integrations, achieving a seamless online‑offline fusion.

Store Digitalization Solutions

Case 1: Convenience Store Beverage Selection

In the BIU pilot store, analysis revealed a 15% gap between female foot‑traffic and beverage purchase rates. By introducing low‑sugar, low‑calorie, and trending drinks favored by women, sales of beverages rose by 20%, demonstrating AI‑driven assortment optimization.

Case 2: Large‑Scale Mall Scene Monitoring

At Suning Life Plaza Ciyun Temple store (47,000 m²), Suning deployed high‑density cameras at entrances, escalator areas, and open corridors, combined with facial capture and pedestrian tracking algorithms. The solution integrates existing surveillance infrastructure with additional edge compute boxes, reducing hardware costs compared to a GPU‑server‑only approach.

These examples illustrate how unmanned store technologies can be scaled to various retail scenarios, enhancing shopper experience, operational efficiency, and data‑driven decision making.

Thank you for listening.