AI Xiaozhao — China Merchants Bank
A conversational AI service layer for enterprise banking, designed to turn user intent into structured task cards, guided workflows, and business actions across web and mobile banking.
- Role
- UX Designer · AI Interaction
- Team
- CMB Network Technology · Experience Design
- Timeline
- 2025–2026 · Enterprise design work and AI-native rebuild
- Outcome
- AI task cards, guided workflows and a reusable conversation system
Original work and AI-native rebuild
This case combines two stages: enterprise UX design work from 2025, and a 2026 AI-native reconstruction built as an interactive coded case study.
The rebuild turned a past enterprise project into the coded case study you are viewing now — reorganising the product logic and rebuilding key conversation flows as clickable components, so readers can explore states, interactions, and feedback directly.
- 2025 · Enterprise UX design
- Owned the task-oriented conversational UX, from intent recognition to task completion
- Defined reusable card patterns, interaction states, and cross-platform handoffs
- Worked with product and developers to translate requirements into implementation-ready flows
- 2026 · AI-native reconstruction
- Reframed the original project into a portfolio-safe interactive case study
- Rebuilt key interactions as clickable components with coding agents
- Reorganised the work into reusable conversation patterns and system rules
Context
In enterprise banking, the users are employees of China Merchants Bank’s corporate clients, for example someone making a payment to a supplier. They do not only ask questions. They need to complete regulated, multi-step tasks across web banking, mobile apps, permissions, forms, images, and business rules. AI Xiaozhao was designed as a conversational service layer that recognises intent, structures information, guides users through business actions, and connects AI assistance with existing banking workflows.
A simple chatbot response was not enough for enterprise banking scenarios. Users often needed to confirm permissions, fill in missing fields, upload materials, read system results, and move between U-Bank, mobile app, and external configuration pages.
The design challenge was to make AI assistance actionable: turning user intent into structured cards, guided steps, and task-specific business flows.
AI Xiaozhao
Design challenge
From answer to action, the challenge was to make AI assistance actionable — recognising intent, structuring the response, guiding the next step, and handing the task back to the right banking workflow.
The table shows how AI Xiaozhao shifted from answering questions to supporting regulated banking tasks.
| Shift | Before | With AI Xiaozhao |
|---|---|---|
| Intent | Users had to find where to start. | AI recognises the task type. |
| Structure | Answers were text-heavy and easy to lose. | Responses become structured, actionable cards. |
| Handoff | The conversation could end before task completion. | The task moves into the appropriate banking workflow. |
Card system
AI Xiaozhao's responses were structured into reusable card patterns for intent recognition, form assistance, query results, guided flows, status exceptions, and handoff actions.
The explorer below turns those patterns into an interactive demo. Try it: ❶ Select a capability tab → ❷ Switch platform context → ❸ Try a preset request. Watch how the AI response and design rationale change together.
We found key invoice fields and suggested next actions.
- Purpose
- Turn ambiguous user intent into immediate next actions, not a generic chatbot reply.
- Used when
- A user asks a broad question, uploads a document, or describes a task that could lead to several possible banking actions.
- Design rule
- Lead with what the user can do next. Show action options first, and keep recognised information as supporting evidence.
Design impact
- 01Reusable interaction patterns
Established a coherent interaction system across query, form, image, result, confirmation, permission, and exception states.
- 02Scalable scenario design
Turned recurring banking tasks into reusable card and interaction patterns rather than isolated screens.
- 03Conversation-to-action handoff
Defined how conversational intent moves into formal banking workflows while preserving task context.
The project shifted AI Xiaozhao from a support chatbot toward a task-oriented service layer inside enterprise banking.
Reflection
This project taught me that enterprise AI design is not about making a chatbot feel smarter. It is about deciding where AI should explain, where it should structure information, where it should ask for confirmation, and where it must hand over to formal business systems.
At the same time, structured cards cannot simplify everything. In banking, some tasks require complete information, full records, explicit terms, permission checks, or human confirmation. A card can reduce cognitive load, but it should not hide complexity that matters for safety, trust, compliance, or accountability.
In regulated banking scenarios, AI Xiaozhao cannot simply 'complete everything'. Its value is to help users understand what is happening, recover from missing information, and move from conversation into the right workflow — without pretending to replace the formal systems behind it.