Qwen App
GenAI shopping agent that searches, compares and places orders in a marketplace catalog from a chat interface
According to its results filed with the SEC, Alibaba counts 250 million users who made their first AI-driven purchase through the agentic features of the Qwen app (up from about 140 million at the end of February 2026), without publishing any effect on conversion or GMV.
Key points
- Alibaba connects its Qwen AI app to Taobao, Tmall and fast delivery so users can shop in conversation.
- Qwen agents with order, logistics and customer service skills, across more than 4 billion products.
- 250 million users have made a first AI-driven purchase through Qwen.
- Evidence A on scale only: no effect on conversion or GMV published, launch heavily subsidized.
Objective
Make Alibaba's consumer AI assistant a new entry point to its marketplaces, by moving the start of the purchase from the search bar to the conversation, and thereby extend the customer reach of the group's e-commerce.
The deployment
Qwen, Alibaba's consumer AI app, was connected on January 15, 2026 to the group's services (Taobao and Tmall, Taobao Instant Commerce, Amap, Fliggy, Alipay) to carry out real tasks: ordering a meal or groceries, booking a ticket, buying a product. The launch was driven by a Chinese New Year campaign starting February 6, with three billion yuan in incentives announced by the group. According to the quarterly results published in March 2026, about 140 million users had made their first AI-driven purchase through Qwen's agentic features by the end of February, and Qwen had passed 300 million monthly active users. On May 11, 2026, Alibaba opened the entire Taobao and Tmall catalog (more than 4 billion products) to the Qwen app, whereas the commerce features had until then covered only a few test categories, and launched a Qwen Shopping Assistant in the Taobao app, reachable from the messages tab: comparisons, virtual try-on for clothing, coupon stacking during shopping festivals, and 30-day price tracking that places the order automatically once the target price is reached. In its results for the quarter ended June 30, 2026, Alibaba reports 250 million users who have made their first AI-driven purchase through Qwen since its launch. The group, however, publishes no conversion, basket or GMV figure attributed to these agents; on the contrary, it cites a rise in inference costs tied to the Qwen app among the losses of its AI segment. The evidence therefore covers adoption at scale, not an isolated commercial effect, and that adoption was obtained with heavy subsidies.
Results Proof A
The adoption figures appear in two earnings releases filed with the SEC (6-K); they prove scale, but no effect on conversion or GMV is published, and initial adoption was bought through a three billion yuan campaign.
How it works
Documented architectureThe stack in detail
- llm Qwen (famille de modeles de langage et multimodaux) In-house foundation model, described as fine-tuned on more than twenty years of transaction and merchant operations data; multimodal model for virtual try-on.
- outil Agents IA a bibliotheque de competences Order management, logistics and after-sales service skills connected to the catalog.
- plateforme Taobao, Tmall et Taobao Instant Commerce The group's marketplaces and fast delivery service, integrated into the Qwen app.
How it runs, concretely
For ops teams-
1Capture the intent AI
The user describes a need in the chat; the agent asks follow-up questions about budget or style.
-
2Search and compare AI
The agent queries the Taobao and Tmall catalog, suggests options or a product bundle and side-by-side comparisons, drawing on the order history.
-
3Order in the chat AI / user
The transaction is completed in the conversation window, with platform coupons stacked; price tracking can trigger the purchase automatically at the threshold set by the user.
-
4Track and provide after-sales service AI / e-commerce platform
The agent's skills cover order management, logistics and after-sales service.
The intent expressed in natural language and the user's order history. Without access to the full catalog and to the order and logistics skills, the agent can only advise and sends the user back to classic search.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- A structured product catalog that can be exposed to an agent (attributes, stock, prices, lead times)
- Order history and customer reviews usable for personalization
- Order, logistics tracking and after-sales APIs the agent can call
Org prerequisites
- A clear decision on how the classic search journey and the conversational journey coexist
- Guardrails on purchases triggered by the agent (confirmation, caps, cancellation)
- Measurement of the incremental effect planned from the start, separate from adoption
Possible stack
- General-purpose LLM with tool calling (function calling) connected to the commerce APIs
- Agentic purchase protocols offered by platforms and payment providers
- Existing product search engine exposed as a tool for the agent
The plan, step by step
- Step 1Expose the catalog, the cart and order tracking as tools an agent can call.Deliverable: A documented and secured commerce API layer for the agent
- Step 2Launch a shopping assistant on a few test categories, with follow-up questions and comparisons.Deliverable: Assistant in production on a limited scope
- Step 3Run a controlled test against the classic search journey and read conversion, basket and returns.Deliverable: Measurement of the incremental effect, separated from volumes driven by incentives
- Step 4Extend to the full catalog and add after-sales skills and price tracking.Deliverable: Agent covering the whole journey, from inspiration to after-sales
First step: Pick a category where comparison is tedious (equipment, home appliances), connect an agent there that queries the catalog and prepares the cart, and compare its conversion with classic search on a randomly drawn sample.
Sources
- S1 Alibaba Group Announces June Quarter 2026 Results (Form 6-K, Exhibit 99.1) Primary archive pending
- S2 Alibaba Group Announces December Quarter 2025 Results (Form 6-K, Exhibit 99.1) Primary archive pending
- S3 Alibaba Opens All of Taobao to Qwen AI, Ushering in a New Agentic Shopping Experience - Alibaba Group Primary archive pending
- S4 Alibaba in 3b yuan Lunar New Year red envelopes push - RTHK Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.