How an AI Agent Completed a Physical Store Order in Hong Kong
AI agents are not limited to online product search. They can also help customers interact with physical stores and local businesses.
DeepLumen recently tested an early version of Agentic Store with WUXIAN TEA in Hong Kong, where an AI agent turned a WhatsApp request into an order that a physical store could fulfill.
TL;DR
- The test took place at WUXIAN TEA in Hong Kong.
- A customer made a request through WhatsApp.
- The AI agent interpreted the request using the store's menu information.
- The agent selected an item and submitted the order to the store.
- Store staff prepared the order through the normal fulfillment process.
- The customer did not need to open the merchant's website or browse the menu separately.
A Small Order With a Larger Question
The order itself was simple: a cup of milk tea.
But the product was not the main point of the test.
The larger question was whether an AI agent could take a natural-language request from a customer and turn it into an order that a real store could fulfill.
During the test, the customer used WhatsApp to communicate a request. The AI agent then interpreted that request, accessed the store's menu information, selected an item, and submitted the order to WUXIAN TEA.
The store's staff prepared the order through the normal fulfillment process.
This was DeepLumen's first use of Agentic Store in a physical retail environment.
Why the Conversational Interface Matters
Traditional ecommerce usually asks customers to follow a fixed path:
- Open a website
- Browse product pages
- Search or filter options
- Add an item to a cart
- Complete checkout
That process works well when a customer already knows how to navigate a store.
But many everyday purchases begin with a simple request rather than a product search.
A customer may say:
- "I want something refreshing."
- "What can I order from this store?"
- "Can you help me find an option that matches my preference?"
In these situations, a conversational interface can be closer to how people naturally express intent.
The customer describes what they need first. The AI agent then uses the merchant's information to identify a suitable option.
The Menu Is Part of the Commerce Infrastructure
For an AI agent to complete an order, the menu must be more than a visual page designed for human browsing.
It must also contain information that a system can interpret accurately.
That includes:
- Available products
- Product names and descriptions
- Relevant options
- Current availability
- Rules that affect the order
If a menu is incomplete or outdated, an AI agent may not be able to respond accurately.
The WUXIAN TEA test demonstrates why structured and current merchant information is important. The agent used the store's menu to understand the available options and select an item before submitting the order.
The Store's Fulfillment Process Stayed the Same
Another important part of the test is what did not change.
Once the order was submitted, the store's staff continued to prepare it through the normal fulfillment process.
The AI agent handled the connection between the customer's request and the order submission. The physical store remained responsible for preparing and fulfilling the order.
For merchants, this creates a more practical path toward Agentic Commerce. Supporting AI transactions does not necessarily require rebuilding the entire store operation.
Instead, an AI agent can connect to the product, menu, and order systems that a merchant already uses.
Beyond Ecommerce
The WUXIAN TEA test shows that Agentic Commerce can extend beyond traditional online shopping.
Potential applications include:
- Restaurants and cafes
- Local retail stores
- Service businesses
- Businesses that receive customer requests through messaging platforms
These businesses often receive requests in natural language. Customers may not want to browse a large catalog or navigate several pages before making a decision.
They may simply want to describe what they need and receive a useful response.
What This Test Demonstrates
The transaction can be summarized as:
Customer request -> AI interpretation -> Menu matching -> Product selection -> Order submission -> Store fulfillment
Previously, an AI system might have told a customer where to find a product.
This test shows a further possibility: an AI agent can participate in the process of turning a customer request into a real order.
FAQ
Did the customer need to visit the WUXIAN TEA website?
No. The customer made the request through WhatsApp and did not need to browse the merchant's website separately.
Which payment method was used?
The payment method was not part of the published case details, so it should not be assumed.
Was this a full public rollout?
The WUXIAN TEA activity was an early Agentic Store test in a physical retail environment.
Did the AI agent replace the store staff?
No. The agent interpreted the request, selected an item, and submitted the order. Store staff continued to prepare the order through the normal process.
Conclusion
A cup of milk tea may look like a small transaction, but it demonstrates an important shift in how customers and merchants can interact.
When a customer can express a request through a conversational interface, and an AI agent can connect that request to a merchant's menu and order system, the customer journey no longer has to begin with opening a website.
For DeepLumen, the WUXIAN TEA test was an early example of Agentic Store operating in a physical retail environment and a starting point for exploring more local retail and service use cases.
See how DeepLumen connects customer requests with real merchant transactions.