A 60-cover restaurant in Lisbon misses an average of 14 phone calls a day during service. Half of those are reservation attempts. At an average ticket of €38 per cover and a typical party of 2.4, that is roughly €1,275 of lost revenue every service. Per location. Per day.
The cause is not bad management. It is physics. A host taking a phone call cannot seat guests, run a POS, or coordinate a turn. So during the hours people actually want to book, the phone goes unanswered.
This is the problem AI agents solve first in restaurants — and it is also the problem with the cleanest ROI.
Where AI fits in a restaurant operation
Think of the restaurant as three operations:
- Front-of-house intake — reservations, takeout orders, questions about the menu, allergies, hours.
- Back-of-house coordination — supplier orders, deliveries, staff scheduling, shift swaps.
- The kitchen itself — cooking the food.
Only the first two are addressable by an AI agent today. The kitchen is not changing. A good AI deployment makes that explicit and stops there.
The reservation problem, solved
The single most-deployed workflow in restaurant AI is reservations on WhatsApp and voice.
A customer messages or calls. The agent reads the live calendar in your reservation system, suggests slots, captures party size and any allergies, takes the deposit if your policy requires one, and sends the confirmation. The host never picks up the phone.
For a 5-location operator, this single workflow typically delivers:
The agent is not a chatbot — see AI agents vs chatbots for the distinction. It reads your real calendar and writes real bookings.
Takeout, allergies, and the long tail
The second workflow most restaurants ship is takeout intake. The agent quotes from your live menu, applies the correct VAT per country, captures allergies (and refuses to take an order if the allergy cannot be safely served), and emits the order to the kitchen.
This is where the conversational ops pattern earns its keep: one thread, one customer, real outcomes.
The long tail is everything else customers ask — "do you have a vegan option," "is there parking," "can we book the back room for 12." These are answered by the agent from your menu, your policies, and your location data. No human is paged unless the question falls outside what the agent is allowed to commit to.
Suppliers and the back office
Less visible but often more profitable: the agent handles supplier coordination.
- Standing orders go out on the right day of the week.
- Delivery confirmations are matched against the order and any discrepancies are flagged.
- Invoices are reconciled against deliveries before they are paid.
A 5-location group typically has one person spending 8-12 hours a week on this. The agent takes most of it.
Deployment: the first 30 days
The pattern that consistently works in restaurants is one workflow, one location, one shift — then expand.
Deploy an AI agent in a restaurant in 30 days
A sequenced rollout from contract to multi-location steady state.
~P30D
- 1
Audit your reservation funnel
Count missed calls per service for one week. This is your baseline and your business case.
- 2
Connect the reservation system
BIOS reads your existing calendar (TheFork, OpenTable, internal). No data migration required.
- 3
Configure policy
Set deposit rules, cancellation windows, allergen handling, and the conditions under which the agent must escalate to a human.
- 4
Shadow week
Agent drafts every reply; a manager approves before send. You discover the 10% of edge cases that matter.
- 5
Go live, one location, one channel
WhatsApp first, voice second. Single location for two weeks. Measure booking conversion and customer NPS.
- 6
Expand to all locations and add takeout
Only after the first location is boring.
What this is not
A few things AI agents do not do in a restaurant, and you should be skeptical of any vendor claiming otherwise:
- They do not replace your host on the floor. Greeting guests, reading the room, fixing problems in person — that is still a human job and will be for the foreseeable future.
- They do not predict demand or set prices. Forecasting is a separate, harder problem.
- They do not write your menu. A model can suggest copy; humans still decide what to cook and what to charge.
The agent is a competent front desk and a competent back-office assistant. That is enough to change the economics of running a restaurant — without pretending to do more.
We hired three people last year to answer the phone and they still missed half the calls. The agent answered 100% in the first week and the calendar filled up. The phone bill went down, the wage bill stayed flat, the bookings went up.
Where the savings actually land
The honest accounting: the agent rarely lets you cut staff. What it does is let the staff you have do the work that humans should be doing — greeting, serving, cooking — instead of triaging the phone.
That shift shows up as: more covers per service, lower no-show rate (deposits + reminders), faster supplier reconciliation, and a measurable drop in manager hours spent answering "what time do you close on Sunday."
Frequently asked questions
What reservation systems does it work with?
Nordix BIOS reads from TheFork, OpenTable, Resy, SevenRooms, and most custom systems via API. If your system has a calendar, BIOS can read it.
What about allergies?
Allergens are first-class in the agent. If a customer mentions an allergen that your menu cannot safely accommodate, the agent refuses the booking or escalates to a human. The decision is logged.
Does it work on phone calls too, or just WhatsApp?
Both. The voice channel handles inbound calls in the same languages as WhatsApp and can transfer to a human when policy requires.
What languages does it work in?
English, Portuguese, Spanish, and Norwegian out of the box. Adding a language is a configuration change, typically delivered in under a week.
Our POS is old. Will this still work?
Probably. BIOS integrates with most POS systems via API or, for legacy systems, via a small connector. The reservation workflow rarely requires POS integration; takeout does.
How is it priced?
Per location, with usage tiers for high-volume operations. Most restaurant deployments pay back within the first month from reclaimed reservations alone.
Stop missing reservations during service.
Nordix BIOS answers every call and message, reads your real calendar, and ships in under two weeks.
Glass preview
Your first operating thread
Real chat · opens in WhatsApp