A customer in a stalled car used to make ten phone calls and wait. Now they say one sentence and roadside rescue is dispatched. Behind that sentence is an insurer serving 251 million people, quietly turning customer service into software. Ping An's "one-sentence service" is one of the clearest pictures of enterprise AI landing in China this year — not a demo, but a default.
The bet: one sentence, three hundred services
In its "Service Year 2026" push, Ping An built an AI assistant it calls Express Service (快捷服务) — the idea is that a customer types or says one sentence and the system understands intent, pulls in the right account, and completes the task. It wraps more than 300 digital services: transactions, financing, claims, emergency rescue.
The reach is the point. Ping An reports roughly 251 million customers across insurance, banking, health and elderly-care businesses, and the assistant is designed to sit in front of all of them rather than inside one app. The framing Ping An uses is "three ones": one sentence to get a service done, one button for an emergency, and a dignity-of-life service for aging customers.
The healthcare twin: an AI doctor at 90 million MAUs
The same strategy runs through medicine. Ping An Good Doctor (1833.HK) upgraded its AI doctor in June 2026 and said the service now reaches about 90 million monthly active users across the group's apps — insurance, banking, auto-owner and health portals. Under the hood is Ping An's medical large model (大模型), "Ping An Medical Master" (平安医博通), trained on the group's medical and insurance data.
In its first-half 2026 interim results (filed August 2026), Ping An reported the AI doctor had served more than 9.7 million users, could precisely diagnose over 11,300 diseases, and matched human-doctor diagnosis at about 96% accuracy. The "AI + human doctor" model now covers 100% of the group's personal customers. Every AI suggestion is reviewed by a licensed physician, with hard cases routed to a person.
Where the money actually shows
Insurance is where the savings are loudest. Ping An said its intelligent anti-fraud systems in property insurance intercepted and avoided losses of about RMB 10.51 billion (≈ US$1.48B / HK$11.5B) in 2025. That is the kind of number a board notices: AI not as a feature, but as a control system on the P&L.
The model also reaches into operations most customers never see — underwriting, second-level approvals, agent assistance — where Ping An reports large efficiency gains as more processes move through AI. Those internal wins are the pump; the customer-facing assistant is the tap.
What "landing" really means here
Two things separate this from a press-release AI. First, it is the default path, not an opt-in chatbot — customers are routed through it for real tasks. Second, it is bounded: the system does the repetitive, rules-heavy work and hands judgement to humans, especially in medicine. That boundary is why a 96% accuracy claim is usable rather than reckless.
The limits are real too. An assistant that "understands intent" still depends on the quality of the backend services it calls; a one-sentence request that needs a human decision is only as fast as that human. And the 251-million-customer scale cuts both ways — a wrong default at that volume is a 251-million-customer problem.
What readers can do now
- If you run a service business: copy the "one sentence → task" shape, but invest more in the backend actions than the chat UI. The assistant is only as good as what it can trigger.
- If you build healthcare AI: take the human-in-the-loop clause seriously — Ping An's 100% coverage and 96% figure rest on physician review, not on the model alone.
- If you watch Chinese AI: track token consumption and anti-fraud savings in earnings calls; that is where "AI landing" turns into audited numbers.
Honest limitations
The "one sentence" assistant, 251 million customers, 300+ services, and the three-ones framework come from Ping An's announcements carried by Xinhua and Securities Times in April 2026. The 90-million MAU AI doctor figure is from Ping An Good Doctor's June 2026 press release (PR Newswire) and subsequent trade coverage. The 9.7 million users, 11,300+ diseases, ~96% accuracy, and 100% personal-customer coverage are from Ping An Good Doctor's H1 2026 interim results announcement (HKEX filing, August 2026). The RMB 10.51 billion anti-fraud saving is a Ping An figure reported by Xinhua and Securities Times for full-year 2025; we converted at roughly 7.1 RMB/USD and 0.915 RMB/HKD and could not independently audit it. Customer and MAU counts are Ping An's own; accuracy and disease-coverage claims are company-reported and medically bounded by human review. We found no independent benchmark of end-to-end task-completion rate at writing.