Starting point
Guest communication & telephony
AI phone assistant for bookings and orders
Every call is answered, in the middle of service and long after closing time: the voice agent books tables, records takeaway orders and answers questions about the menu from the restaurant's own maintained content. Kitchen, booking book and guest receive the same information in the same minute.
Process architecture
From the ringing phone to a confirmed booking or order.
The chain connects the phone system with the booking book, the point-of-sale system and the maintained knowledge about the menu and the house. Every conversation ends either with a created record or with a note for the service team, never with a lost call.
01Input
Call & context
Calls and messages enter one case. Guest profile, history and the verified knowledge about menu and house are ready before the first sentence.
4 Steps
02Process
Conversation & checks
The voice agent clarifies the request, answers questions from the menu, checks capacity and rules and creates the record. Anything unclear goes to the service team.
8 Steps
03Output
Confirmation & record
Guest, kitchen and service receive the same information, the conversation is logged and open questions feed back into maintaining the knowledge base.
4 Steps
Source01Answer the call
The SIP trunk of the phone system routes calls outside service hours and on a busy line to the voice agent, which picks up after two rings.Details
Answer the call
The SIP trunk of the phone system routes calls outside service hours and on a busy line to the voice agent, which picks up after two rings.
Interfaces
- Microsoft Graph API
- SharePoint API
Next steps
- Identify the guest
Source02Pick up messages
Messages from the WhatsApp Business Platform and the chat on the restaurant website enter the same case, so guest and request are handled only once.Details
Pick up messages
Messages from the WhatsApp Business Platform and the chat on the restaurant website enter the same case, so guest and request are handled only once.
Interfaces
- Microsoft Graph API
- SharePoint API
Next steps
- Identify the guest
Data03Identify the guest
The Nexaluna Agent API loads the caller number, previous bookings, table preferences and recorded allergies whenever the guest is already known.Details
Identify the guest
The Nexaluna Agent API loads the caller number, previous bookings, table preferences and recorded allergies whenever the guest is already known.
Interfaces
- Azure AI Document Intelligence
- Mistral OCR
- Dataverse API
Next steps
- Load the knowledge base
Data04Load the knowledge base
Menu, daily specials, opening hours and the website copy are stored as searchable sections in Qdrant and serve as the verified source during the call.Details
Load the knowledge base
Menu, daily specials, opening hours and the website copy are stored as searchable sections in Qdrant and serve as the verified source during the call.
Interfaces
- Azure AI Document Intelligence
- Mistral OCR
- Dataverse API
Next steps
- Hold the conversation
Action05Hold the conversation
ElevenLabs Conversational AI speaks with the guest in natural language, asks for missing details such as date, party size or pickup time, and keeps the state.Details
Hold the conversation
ElevenLabs Conversational AI speaks with the guest in natural language, asks for missing details such as date, party size or pickup time, and keeps the state.
Interfaces
- Power Automate
- Azure Functions
- Microsoft Graph API
Next steps
- Classify the request
- Service steps inGuest wants a person
Decision06Classify the request
The conversation is assigned to one of three tracks: information about the menu and the house, a table booking, or an order for pickup or delivery.Details
Classify the request
The conversation is assigned to one of three tracks: information about the menu and the house, a table booking, or an order for pickup or delivery.
Interfaces
- Azure OpenAI GPT
- Anthropic Claude
- AI Builder
Next steps
- Answer questionsInformation
- Check table availabilityTable booking
- Take the orderOrder
Action07Answer questions
Questions on dishes, allergens, opening hours and directions are answered from the stored knowledge base; without a supported passage the agent says so.Details
Answer questions
Questions on dishes, allergens, opening hours and directions are answered from the stored knowledge base; without a supported passage the agent says so.
Interfaces
- Power Automate
- Azure Functions
- Microsoft Graph API
Next steps
- Log the conversation
- Review knowledge gaps
Action08Check table availability
Free capacity for the requested slot is queried through the REST API of the booking system, and a nearby alternative is offered where needed.Details
Check table availability
Free capacity for the requested slot is queried through the REST API of the booking system, and a nearby alternative is offered where needed.
Interfaces
- Power Automate
- Azure Functions
- Microsoft Graph API
Next steps
- Check feasibility
Action09Take the order
Dishes are resolved against the current menu, extra requests and allergens are noted, the price is totalled and the order is read back to the guest.Details
Take the order
Dishes are resolved against the current menu, extra requests and allergens are noted, the price is totalled and the order is read back to the guest.
Interfaces
- Power Automate
- Azure Functions
- Microsoft Graph API
Next steps
- Check feasibility
Decision10Check feasibility
Kitchen and dining room capacity, lead time and minimum order value are checked against the stored rules before anything is confirmed as binding.Details
Check feasibility
Kitchen and dining room capacity, lead time and minimum order value are checked against the stored rules before anything is confirmed as binding.
Interfaces
- Azure OpenAI GPT
- Anthropic Claude
- AI Builder
Next steps
- Create the record
- Service steps inException
Approval11Service steps in
Large parties, events, complaints and every explicit request for a person are handed to the service team together with the full conversation note.Details
Service steps in
Large parties, events, complaints and every explicit request for a person are handed to the service team together with the full conversation note.
Interfaces
- Teams Adaptive Cards
- Power Apps
Next steps
- Create the recordAfter approval
System12Create the record
Booking or order is written through the REST API into the reservation and point-of-sale system, with a key that prevents a duplicate on a dropped connection.Details
Create the record
Booking or order is written through the REST API into the reservation and point-of-sale system, with a key that prevents a duplicate on a dropped connection.
Interfaces
- Dynamics 365 API
- Azure Service Bus
Next steps
- Confirm to the guest
- Inform the operation
Result13Confirm to the guest
The guest receives the confirmation with time, party size or order contents by text message, email or on the same channel they used to get in touch.Details
Confirm to the guest
The guest receives the confirmation with time, party size or order contents by text message, email or on the same channel they used to get in touch.
Interfaces
- Power BI API
- Teams Webhook
Next steps
- Log the conversation
Result14Inform the operation
Kitchen and service receive the order as a ticket and the booking in the daily overview, without anyone retyping a handwritten note.Details
Inform the operation
Kitchen and service receive the order as a ticket and the booking in the daily overview, without anyone retyping a handwritten note.
Interfaces
- Power BI API
- Teams Webhook
Result15Log the conversation
Summary, time, channel and the decision taken are stored on the guest record so that every step remains traceable afterwards.Details
Log the conversation
Summary, time, channel and the decision taken are stored on the guest record so that every step remains traceable afterwards.
Interfaces
- Power BI API
- Teams Webhook
Next steps
- Review knowledge gaps
Result16Review knowledge gaps
Questions without a supported answer are collected in a weekly overview and used to sharpen the menu, the website copy and the agent's rules.Details
Review knowledge gaps
Questions without a supported answer are collected in a weekly overview and used to sharpen the menu, the website copy and the agent's rules.
Interfaces
- Power BI API
- Teams Webhook
System landscape
Phone system, booking book and menu working together.
- Microsoft
Microsoft Graph API
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
SharePoint API
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Azure AI Document Intelligence
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Mistral OCR
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Dataverse API
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Power Automate
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Azure Functions
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Azure OpenAI GPT
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Anthropic Claude
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
AI Builder
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Teams Adaptive Cards
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Power Apps
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Dynamics 365 API
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Azure Service Bus
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Power BI API
Möglicher Baustein für die Microsoft-Systemvariante.
- Microsoft
Teams Webhook
Möglicher Baustein für die Microsoft-Systemvariante.
- Interfaces in use
SIP-Trunk
Answers the call through the existing phone system and hands it to the voice agent without a new number.
- Interfaces in use
ElevenLabs Conversational AI
Handles the conversation in natural language, recognises the request and asks for missing details.
- Interfaces in use
Nexaluna Agent API
Finds the guest, previous bookings and open orders during the call and writes the note back.
- Interfaces in use
Qdrant
Holds the menu, daily specials and website copy as a searchable base from which answers are supported.
- Interfaces in use
REST API
Creates bookings and orders in the reservation and point-of-sale system and queries free capacity.
- Interfaces in use
WhatsApp Business Platform
Receives messages on the channel many guests already use and sends the confirmation back there.
- Possible apps & platforms
OpenTable
Booking book that the service team keeps using unchanged, extended by the bookings made in the conversation.
- Possible apps & platforms
resmio
Alternative for restaurants that already keep their booking book here, with the same capacity and lead-time rules.
- Possible apps & platforms
Lightspeed Restaurant
Point-of-sale system where orders arrive as a kitchen ticket and are settled.
- Possible apps & platforms
WhatsApp Business
Message channel for guests who prefer writing to calling, with the same information as on the phone.
- Possible apps & platforms
Nexaluna Agent Console
Interface for the operation: conversation histories, exceptions awaiting approval and maintenance of the knowledge base.
Implementation in context
Robust during live service.
The process replaces no system, it connects the existing ones: the phone number stays unchanged, the booking book remains the system of record for tables, the point-of-sale system remains it for orders. The voice agent is the additional channel in front of them, not another program beside them.
Answers about the menu come from a maintained base only: menu, daily specials, opening hours and the website copy are stored as searchable sections. If the agent finds no supported passage, it says so and offers a call back rather than inventing an ingredient or an allergen.
The limits are part of the setup: up to which party size a booking is made automatically, from when a deposit is required, how much lead time the kitchen needs for a pickup and at which times nothing is accepted at all. Anything beyond that goes to the service team with the full conversation history.
Personal data is involved: name, phone number, occasionally a note about an intolerance. The place of processing is defined, the notice about recording is given at the start of the conversation, and a fixed retention period applies to recordings and logs.
Target picture
Calls and messages are answered around the clock and lead to an entry in the booking book, an order in the point-of-sale system or an answer supported by the restaurant's own content. The service team only steps in where a decision of the house is required, and sees every case with a full note.
Expected benefits
- Calls are answered in the middle of service and after closing time instead of going unanswered.
- Bookings appear in the booking book straight away, without anyone copying a handwritten note.
- Recurring questions about the menu, allergens and opening hours no longer occupy experienced service staff.
- Takeaway orders reach the kitchen as a ticket carrying the same details the guest heard on the call.
- The service team keeps the decision on groups, events and special requests and receives the full note for it.
- Open questions from the conversations show where menu and website are unclear and can be improved deliberately.
Prerequisites to start
- 01A phone system with a SIP trunk, or a provider that allows the number to be forwarded to the agent.
- 02A reservation system with a documented interface and a technical account with write permissions.
- 03A maintained, current menu with prices, additives and allergens in machine-readable form.
- 04Defined rules for capacity, party size, lead time and minimum order value.
- 05A named person per shift who takes over the exceptions raised by the agent.
- 06A data processing agreement, a notice about recording and a retention period for recordings and logs.
Your next step
Which process costs your team unnecessary time every day?
Give us a short outline. We will tell you honestly where automation is useful—and where it is not.
