AI reservations agent: what it takes over and what stays human

What a reservations agent actually does all day

In most restaurants and cafés, the reservations agent is the person who answers the phone, watches the online book and keeps each service in a shape the kitchen can handle. The job title varies. In some venues it is a host, a booking coordinator or a front-of-house manager who does this on top of other duties. The work itself repeats across all of them.

Most days include a few predictable jobs. A booking comes in by phone, WhatsApp, Instagram direct message, email or a form on the restaurant's Google listing, and the agent checks whether there is a table for that party size at that time. Once a slot is agreed, the agent sends a confirmation. The day before service, the agent chases bookings that have not confirmed. When a guest cancels, the agent releases the table and, if there is a waitlist, offers it to the next party. When a guest calls to say they are bringing two more people, or two fewer, the agent updates the party size and checks the floor plan again. Special requests go onto the booking: a birthday, a nut allergy, a high chair, a request for the window. Large groups get passed to a manager who decides whether to ask for a deposit. After service, someone marks which bookings never arrived. In between, the agent answers questions such as whether there is anything at seven, and takes the occasional complaint about a long wait or a booking that was lost.

Those are the core tasks. What differs between venues is the volume and the channels. A fine dining room may handle fewer calls but spend much longer on each one. A café may take a constant stream of quick requests for weekend brunch. The agent in each case is judged on the same things: the book is accurate, the phone is answered, and the kitchen is not surprised at 8 p.m.

Which of those tasks an AI worker takes over

An AI worker can take over the tasks that follow a rule, use data that already exists and produce a record that someone can check later. Booking intake is the first. An AI agent can answer web chat, WhatsApp Business messages, and in some setups a phone line, check live availability and write the booking into the reservation system. The restaurant needs its booking tool connected to that agent, usually through the vendor's integration or an import. Common options include Resos, OpenTable, Resy, SevenRooms, Eat App and Tock, and the right choice depends on which of them the restaurant already pays for. Phone calls need a voice line with a call-forwarding setup, and the business must decide what the agent may and may not book on the phone.

Confirmations and reminders are the best-supported part of the job. A reminder can go out by SMS through the booking system's own messaging or through a provider such as Twilio, and by WhatsApp where the restaurant has set up approved message templates. The Resos No-Show Index 2026 reports that reservations with a delivered SMS reminder had a 16% lower recorded no-show rate than the same restaurants' reservations without one, across 46,884 reservations at 257 restaurants. Resos calls this an observed association rather than a controlled test, so treat it as a signal for your own venue, not a guaranteed result. The reminder should ask the guest to confirm or cancel, and the reply should update the booking without a human reading it.

Cancellations and waitlist moves follow the same pattern. When a guest cancels by reply, the agent writes the change to the booking tool and offers the freed slot to the next waitlisted party. The Resos index also shows that 56% of recorded cancellations arrived within 24 hours of the booking or after it had started, so speed matters. An agent that releases a table at 10 p.m. the night before is worth more than one that waits for the morning shift to read a sheet.

Party-size changes are a rules task. The agent checks the new size against the floor plan and the time, then confirms or offers another slot. Deposit flagging works the same way. The restaurant sets a rule, for example that parties of eight or more, or any booking for a private room, are flagged for a deposit. The agent sends the payment request through a payment link from the provider. Card details should stay with that provider, which is the approach the PCI DSS standard expects for card handling. The agent should never store full card numbers in a spreadsheet.

End-of-service reconciliation is the last piece. The agent can export the day's bookings, no-shows and cancellations to Google Sheets or Microsoft Excel and, where the business needs it, pass deposit payments to the accounting tool such as Xero or QuickBooks. Someone still reads the summary. The agent's job is to make the summary accurate and on time.

What stays with a person, and why

A person keeps the tasks that need judgement, a relationship or a signature. Complaints are the clearest case. A guest who waited forty minutes after a booking was lost needs an apology, a decision about compensation and sometimes a manager's voice. An AI agent can log the complaint and route it. It should not decide how to make it right.

Special occasions are close behind. An anniversary proposal, a surprise cake, or a family that has booked the same table every year for a decade calls for someone who knows the room. Regulars and VIP guests also tend to expect a person on the phone. The agent can note the preference, but a human decides who gets the best table on a busy Saturday.

Walk-in judgement calls stay human too. Whether to squeeze in a party of five without a booking, or to hold a table for a late guest who has been delayed on the metro, depends on the floor, the staff and the mood of the room. The same applies to waiving a deposit or a no-show fee. The agent can calculate what the policy says. A manager should decide whether the policy applies this time.

Anything regulated also needs a named person. Deposit refunds and no-show charges sit under consumer law, and the rules differ by country. A person should own the policy wording, approve exceptions and answer any formal request about personal data. In the European Union, Article 50 of the AI Act, which has applied since 2 August 2026, requires that people are told when they are talking to an AI system, at the latest at the first interaction. That duty falls on the business as well as the software vendor, so the agent must introduce itself as automated on every channel where it speaks to guests.

How this role differs by business type

The core tasks look similar across venues, but the emphasis changes. The table below shows six common types of restaurant and café, and what the reservations agent does in each.

In the fine dining and private dining rows, the agent's value sits in the paperwork that surrounds a booking. In the casual and café rows, it sits in speed. In the group row, it sits in consistency. In every row, the person still decides what happens when the rules do not fit the guest in front of them.

How the switch happens, week by week

The switch works best when it is staged. A business that turns everything on at once will find that the first failure is a double booking on a busy Friday. The sequence below assumes one reservations agent on the team, and a booking system that already holds the restaurant's real data.

1. Week 1: Record every booking task for one week. Note the channel, the time taken and the result. This becomes the baseline and shows which tasks repeat most.
2. Week 2: Connect the booking tool and the shared inbox. Switch on confirmations only, and have the agent send them while the person checks a sample each day.
3. Week 3: Add reminders at a set interval before service, with a reply option to confirm or cancel. Keep the person reviewing the no-show log at the end of each service.
4. Week 4: Add cancellations and waitlist moves. Route any reply the agent cannot read, and any complaint, straight to the person on duty.
5. Weeks 5 and 6: Add party-size changes and deposit flags for the rules you have written down. Check every flagged deposit for the first two weeks.
6. Weeks 7 and 8: Review the exception log. Fix the rules that caused most exceptions, then cut the daily check to a weekly sample.
7. After week 8: The person reviews the summary, handles the exceptions and owns the policy. The role has shrunk from booking every table to reviewing what the agent did and deciding the cases it cannot.

The timeline is a planning guide, not a promise. A restaurant with a clean booking history and one channel can move faster. One with three booking tools, a paper book and a group that changes its policy every season will need longer.

Risks, and when not to automate this role

The first risk is a double booking. If the agent writes to one book while a host also writes to a paper list, the room can end up with two parties at one table. Pick one source of truth, and make sure every manual change goes into the same tool the agent reads.

The second risk is an unclear deposit rule. Guests dispute charges they did not understand when they booked. The policy must be shown before the guest completes the booking, and the cancellation cut-off must be written in plain words. In South Korea, the Fair Trade Commission's Consumer Dispute Resolution Criteria, effective 18 December 2025, set higher no-show fee caps for reservation-based restaurants than for general restaurants, and the higher cap applies only where the restaurant disclosed its category and penalty terms in advance. Check the consumer rules where you trade before you copy a policy from elsewhere.

The third risk is that the reminder works too well for the wrong guests. A reminder that goes to a wrong number reaches a stranger. Check that phone numbers are captured in a consistent format, and keep an opt-out path for guests who ask to stop messages. Where guests' personal data is held, the business should know where it is stored and who can see it.

The fourth risk is poor fit. Some businesses should not automate this role yet. A walk-in-only café has no bookings to confirm. A restaurant with the room half empty most nights has a demand problem, and reminders will not fill seats. A venue where the owner takes every booking personally, and where the booking is part of the relationship, may lose something it values if the agent answers first. The Resos index also shows that dinner runs at 2.63% against 1.84% for lunch, which suggests that the late service may need the closest attention. Start with the service that loses the most tables.

Finally, remember what the 2.33% figure is. It is the rate that staff recorded, across restaurants that chose to record no-shows. The Resos calculator page notes that only 72.9% of eligible venues marked at least one no-show, so the true rate may be higher. Measure your own no-show and cancellation rate for four weeks before you switch anything on, so you can see whether the change helped.