Donna
Guest conversations that lead to action.
THE PRODUCT
Guest requests do not follow the hotel's org chart. One call can begin with a room booking, move to a question about the spa and end with a room-service order. Donna maintains one conversation across those services, using the voice, policies and current information of each property. Her work does not stop at recognizing intent. Donna gathers the required details, checks what the hotel can provide and turns supported requests into a booking, an order or an operational task. She confirms important information before acting. When an exception appears, she gives staff the organized context they need to continue without making the guest start again.
IN USE
- (01)
The guest asks in their own words
The conversation can combine a booking, a question about hotel services and a request for the stay.
- (02)
Donna checks and confirms
She keeps the context, consults current hotel information and confirms dates, prices and critical details.
- (03)
The conversation becomes work
Donna completes a supported action or gives staff a structured handoff they can continue without repeating questions.
ENGINEERING
Donna combines a property-specific configuration layer with typed connectors to reservation, service and hotel operations systems. A conversational state machine tracks active intents, required fields, source data, confirmations and completed actions as the call moves between departments. Reads and writes are handled differently: every tool validates its schema before execution, while operations that modify a system retain an idempotency key and commit state to prevent duplicate bookings or orders during retries.
Retrieval is restricted to approved property sources, and availability, prices, opening hours, allergens and terms are checked at the time of the request. Payment moves to a separate secure flow. If a connector times out, returns incomplete information or conflicts with the call context, Donna stops the action and creates a structured handoff containing the collected fields and reason. Evaluations replay multi-intent calls and measure field extraction, action accuracy, tool latency, duplicate prevention and handoff completeness.
