You open your banking app not to tap through menus but to type a sentence: "What dragged my fund down this week, and should I rebalance?" Within seconds you see the loss, the three stocks behind it, and a suggested move — then a button to talk to a human if it gets complicated.
That is the experience China Merchants Bank (招商银行) is building around "AI Xiao Zhao" (AI小招), a smart wealth assistant that has become the face of the bank's "AI First" push.
Key takeaways
- China Merchants Bank (招商银行, HKEX: 3968) launched "AI Xiao Zhao" (AI小招), a smart wealth assistant, in 2023 on a privately deployed base large model (大模型) with a "human + machine" review mechanism.
- In 2025 it open-sourced "Yi Zhao" (一招), described in the bank's annual report as China's first open-sourced 10-billion-parameter financial large model (百亿参数金融大模型), released with a 2TB financial dataset.
- The bank reports more than 120 large-model application scenarios across retail, corporate, risk control and operations.
- Its two consumer apps, the CMB App and the CMB Lifestyle app, reached 123 million monthly active users, according to the bank's 2025 annual report and Euromoney.
- Euromoney reports CMB's proprietary AI model runs into tens of billions of parameters and that its chatbots resolve about 90% of customer-service enquiries without a human.
What the assistant actually does
"AI Xiao Zhao" (AI小招) started as a wealth companion: financial analysis, product selection, market views and return breakdowns, all through conversation. The bank designed it to connect to a remote relationship manager when a question gets too complex, so the AI handles the routine and hands off the edge cases.
Behind it sits a deliberately narrow model. Rather than chase a general large model (大模型), CMB built a finance-specific one, "Yi Zhao" (一招), and open-sourced it in 2025 with a 2TB dataset drawn from more than three decades of the bank's desensitized financial text. Opening the weights was a strategic bet: it lets smaller banks and insurers reuse the model instead of building their own.
The annual report puts the bank's large-model footprint at more than 120 scenarios. A separate internal AI trading bot for interbank markets is reported by the bank to exceed 97% accuracy on its pricing and quoting tasks.
Why a bank is betting its interface on AI
CMB is nicknamed the "retail king" of Chinese banking, and its edge has always been service design, not just balance sheet. Euromoney's 2025 coverage notes the bank grew assets under management about 12% to roughly Rmb 15 trillion (≈ US$2.1 trillion) even as industry margins compressed, with fee income from wealth and lifestyle services reaching 37% of revenue.
The AI layer is the latest version of that strategy. If a chatbot can absorb 90% of routine enquiries, human staff shift to the conversations that actually need judgment. The bank frames this as "human + digital" rather than "human versus machine," and it extends to elderly users through a simplified app version and a screen-sharing voice support feature.
The open question is trust
A financial assistant that can suggest a rebalance is useful only if customers believe the guardrails are real. CMB says it runs a "human + machine" review loop and keeps humans accountable for decisions, a stance echoed across Chinese banks wary of AI giving unsupervised advice.
The broader signal is that consumer banking's front door is becoming a conversation, not a menu tree. For a generation that grew up on messaging, typing a question feels more natural than hunting through tabs — and a bank that answers well keeps the relationship.
The same playbook is spreading internally, too. CMB reports that AI coding assistants now cover the large majority of its developers, and that across the bank the technology replaced a reported 15.56 million hours of manual work in 2025. Whether those internal productivity numbers hold up to scrutiny, the direction is clear: the assistant users see is only the visible tip of an institution being rebuilt around models.
Honest limitations
The headline figures here come from the bank's own annual report, its official statements and Euromoney's award coverage — not from an independent audit of AI performance. The "90% of enquiries resolved" and "123 million MAU" numbers are reported, not verified by a third party.
"First open-sourced 10-billion-parameter financial model" is the bank's self-description of a regulatory and technical milestone; independent confirmation of the parameter count and dataset quality is limited. AI financial advice still sits under human oversight by design, and any suggestion it produces should be treated as guidance, not a recommendation. Deployment depth also varies by product and by whether a given customer is routed to the AI or to staff.
What readers can do now
- If you bank with CMB, open the app and try "AI Xiao Zhao" (AI小招) on a single question — a fund's weekly move — and check whether the explanation matches your own statement.
- Treat any bank AI's rebalance suggestion as a starting point; confirm with a human adviser before moving real money.
- Watch the open-sourced "Yi Zhao" (一招) model: if smaller banks adopt it, AI wealth assistance could become a default feature across the industry rather than a flagship perk.