According to Sina Technology, which obtained information from someone familiar with the matter, Vipshop has recently quietly launched its own AI assistant “AI Xiaowei”. It was developed by the company’s internal large-model team. The assistant supports text and voice conversations and can help users find products, check promotions, and ask about discounts. The e-commerce platform’s ambitions in AI are evident from its recruitment efforts: Sina Technology examined recruitment platforms and found that VIP.com is offering high salaries of 50K to 80K per month, which equals an annual salary of 600,000 to 960,000. The company is seeking top talent for core positions such as large model training, data editing, and SFT text data, aiming to increase the density of skilled personnel.
Product comparison grayscale testing, removing entrance first then changing
The focus of this team is to break down the modules required for “Xiao Wei”’s purchasing intent recognition, product understanding, recommendation generation, and multi-round dialogue, and to continuously add data and improve weaknesses. Another “AI product comparison” function is also being refined: previously, when searching for categories like sports shoes, the detail page would directly display an AI-generated PK module, which automatically extracted selling points and listed parameters for horizontal price comparison; however, the front-end direct display entry has recently been removed and redeveloped. The customer service confirmation tool is still being improved. The new logic now has two options: select a product from the shopping cart and click “Compare”, or directly enter “Xiaowei” to compare products.
Under pressure from performance, AI becomes the growth engine
The pressure behind it is clearly visible: VIP Shop’s net revenue in the second quarter of 2026 was 24.7 billion yuan, a 4.3% decline year-on-year. Chairman Shen Ya calls AI the “key engine” that drives growth and efficiency improvement. At a time when the clearance business is at its peak, it remains to be seen whether self-developed large models can create a new trend.