JD Health
free consumer AI agent that routes the request toward a product purchase and serves as the base for commercialization
JD Health, which runs the largest online medicine sales platform in China, grew the user base of its Dr. Dawei AI health agent nearly fourfold in the first half of 2026. The agent has logged several hundred million interactions and a 98% favorable rating, but its effect on revenue is not quantified publicly.
Key points
- JD Health operates Dr. Dawei, a consumer AI health agent built into its pharmacy e-commerce platform.
- Its user base grew nearly fourfold over the first half of 2026.
- Several hundred million cumulative interactions, 98% satisfaction, nearly 70% of users in tier 3 cities and below.
- Evidence level A: earnings call of August 13, 2026, but the monetization effect stays qualitative, with no revenue figure.
Objective
Convert a free health consultation audience into purchases on JD's pharmacy platform, chaining triage, teleconsultation, testing and medicine purchase inside a single conversation. Management says it then wants to turn this into a business line of its own, with agreements already signed, but publishes no revenue figure attributed to the deployment.
The deployment
JD Health runs the largest online medicine sales platform in China. Dr. Dawei is its consumer-facing doctor agent: the user asks a health question in the app, the agent triages it, runs a consultation-style exchange, reads a test report if needed, then points to a product in the pharmacy catalog or to a doctor at the group's online hospital. The agent was presented on September 25, 2025 among the ten or so agents of AI Hospital 1.0, all built on the in-house medical model Jingyi Qianxun, whose 2.0 version was unveiled the same day. JD Health describes this architecture as a chain running from triage to consultation, testing and medicine purchase. In the first half of 2026, Dr. Dawei gained a long-term health record memory, which supports follow-up between two exchanges instead of a one-off answer. The half-year report states that the number of users served by the agent rose nearly fourfold year over year during the 618 festival, and the earnings call ties that growth to better conversion into product purchases. Cumulatively, the agent has logged several hundred million interactions, a 98% favorable rating, and nearly seven users out of ten coming from tier 3 cities or below. One point stays open: the only figure attached to the deployment is this audience growth. Management says it has signed commercialization agreements for Dr. Dawei and sees a clear path, without publishing revenue, basket size or ARPU. The rise of more than 20% in the number of advertising merchants, cited on the same call, measures JD Health's advertising platform and is not presented as an effect of the agent.
Results Proof A
The central figure comes from the August 13, 2026 earnings call on first-half results, and it is corroborated by two established press outlets relaying the half-year report (Science and Technology Daily, China News Service). Limit to accept: only the agent's audience growth is quantified and attributed to the deployment. The monetization effect stays declarative (commercialization agreements signed, clear path ahead), with no published revenue or basket size, and the rise of more than 20% in advertising merchants measures the advertising platform, not Dr. Dawei. Axis classification: this case is filed under activation_conversion rather than monetization, because the only effect measured and attributed to the deployment concerns audience and routing toward purchase. Monetization of Dr. Dawei is announced, not yet quantified; filing it under monetization would have filled an empty cross-section with an unproven result.
How it works
Documented architectureThe stack in detail
- llm Jingyi Qianxun 2.0 In-house medical model from JD Health, presented in version 2.0 on September 25, 2025. It is the model the AI Hospital 1.0 agents rely on, including the doctor agent Dr. Dawei.
How it runs, concretely
For ops teams-
1Entry through the health question customer
The user describes a symptom or uploads a test report in the JD Health app.
-
2Triage and exchange AI
Dr. Dawei, built on the Jingyi Qianxun model, routes the request, runs the consultation-style exchange and reads the documents provided.
-
3Record recall AI
The health record memory added in the first half of 2026 makes it possible to pick up the user's history from one exchange to the next, instead of starting over.
-
4Move to action AI
The exchange leads to a product in the pharmacy catalog, a test, or a handoff to a doctor at the group's online hospital.
-
5Human relay human
Doctors at the online hospital take over whatever falls outside the agent's scope, in particular anything that requires a prescription.
The link between the health answer and a product available in the pharmacy catalog plus an execution capacity (medicine delivery, testing, a doctor within reach). If the catalog, the stock or the logistics are not wired into the conversation, the agent answers but converts nothing: audience growth then turns into nothing measurable on the commerce side.
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
- Product catalog with stock and availability actually queryable from the conversation
- Persistent user history between two sessions, so the agent follows a record instead of answering case by case
- Reliable domain corpus to ground the answers, with an explicit scope of what the agent refuses to handle
Org prerequisites
- An execution chain behind the conversation (order, delivery, appointment booking), otherwise the agent produces audience and nothing else
- A qualified human relay for cases outside the authorized scope, and the rule that triggers that relay
- A compliance framework validated before going live when the domain is regulated
Possible stack
- Language model specialized on the domain, owned or adapted
- Retrieval layer over the domain corpus and over the live catalog
- Persistent user memory attached to the account
- Gateway to the ordering tool and to the human back office
The plan, step by step
- Step 1Write down what the agent handles and what it passes to a human, and code that scope as a routing rule.Deliverable: Documented scope of intervention and escalation rule.
- Step 2Ground the answers on the domain corpus and on the catalog queried in real time, rather than on the model's knowledge alone.Deliverable: Agent that only recommends products actually available.
- Step 3Add a persistent user memory, to turn the one-off answer into follow-up.Deliverable: History reusable from one session to the next.
- Step 4Instrument the conversation all the way to the action, so conversion can be tied to the agent and not only to overall traffic.Deliverable: Conversion measurement attributed to the agent.
- Step 5Open commercialization only once that measurement is in place, otherwise audience growth stays the only available indicator.Deliverable: Revenue model backed by a measurement, not by usage volume.
First step: Wire the agent to the catalog and to the execution capacity before opening the conversation to the public: measure the share of exchanges that end in an action first, not the volume of exchanges.
Sources
- S1 Earnings call transcript: JD Health posts strong H1 2026 profit growth Secondary archive pending
- S2 京东健康:人工智能驱动诊疗提效,AI医生服务用户增近4倍 (Science and Technology Daily) Established press archive pending
- S3 AI医生进医院,新药首发到基层:京东健康半年报里的普惠医疗实践 (China News Service) Established press archive pending
- S4 2025京东全球科技探索者大会:京东健康发布"AI医院"、升级"京医千询2.0" (Beijing Business Today) 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.