A restaurant booking agent needs to represent the table the restaurant can actually offer. Zanus AI should be evaluated against party size, location, timing and change requests, rather than only whether it can repeat an opening-hours answer during a demonstration.

Confirm which reservation rules are represented

The restaurant page describes booking within party-size rules and routing larger parties or private-room requests to an events manager. The AI Booking page describes service and location availability, calendar connections and booking confirmation. Verify the actual reservation system and rules in the account you intend to use.

Write down table capacity, seating intervals, expected turn time and which requests require staff approval. A party of six at seven o’clock is not interchangeable with two independent tables of three unless the restaurant’s policy and available layout support that arrangement.

Use a reservation record the host can understand

FieldWhat the host needs
LocationThe exact restaurant, not just the group brand
PartyCurrent headcount and relevant seating request
TimeDate, arrival time and any agreed limitation
StatusConfirmed, awaiting action or cancelled

Try a changed headcount

In an illustrative test, book four guests and then ask to increase the party to seven. Inspect whether the original availability is reassessed and whether the record the host sees reflects the outcome. If the larger party cannot be accommodated automatically, the caller should receive an accurate next step instead of an unsupported confirmation.

Repeat with a requested date change. Check the original reservation as well as the proposed replacement. A successful conversation should not leave two live bookings when the customer intended to move one.

Keep special requests distinct from accepted commitments

A requested window table, private room or unusual arrival time may require approval. Record the request clearly and define who can confirm it. Do not make the host discover an unapproved promise when the guest arrives.

For menu and dietary questions, provide only the restaurant’s approved information and a clear staff route for questions that need the kitchen. This article does not assess the system as a substitute for staff making a dietary assurance.

Review the service before a busy sitting

Use internal test callers during a controlled period and have the host inspect the live reservation view. Include a full sitting, an unavailable location, a late change and a cancellation. Review duplicate reservations, unsupported promises and unresolved requests. The useful outcome is a book the front-of-house team can trust throughout service.

Mapping Table Configurations Before the First Call Arrives

Before any AI booking agent handles a live call, the restaurant needs a written inventory of what it can actually seat. This is not simply a count of chairs. It includes which tables can be combined, which cannot, which positions are accessible, and which carry restrictions such as proximity to service stations or exterior doors. A booking agent operating without this map will either confirm arrangements the floor cannot support or refuse requests the floor could accommodate with a small adjustment.

A practical starting point is to walk the dining room during a quiet period and assign each table a realistic range: the minimum party size that makes it worthwhile to hold the table and the maximum that fits safely and comfortably. Record these ranges in the same system the booking agent draws on. If a corner booth seats up to five but the restaurant prefers not to seat fewer than three there on a Friday evening, that preference should be explicit rather than left to front-of-house discretion at the point of arrival.

Once the table inventory exists as a document, compare it against the rules fed to the booking agent. Look for gaps. A table that seats eight might appear in the system as two tables of four that can be joined, or it might appear only as two independent tables. The distinction matters when a caller asks for seven. If the system cannot see the joined configuration, it may confirm two tables of three and four as a workaround, leaving the host to resolve a seating conflict at the door.

Seating intervals are a separate layer. If the restaurant turns tables every ninety minutes and a caller wants a booking at the middle of a service period, the agent needs to know whether that slot is open or whether it falls within a window already committed to a prior party. An AI agent that responds to availability questions without reference to turn time may confirm a slot that is technically open on the calendar but operationally impossible for the kitchen.

Designing a Fictional Enquiry to Evaluate Booking Accuracy

One useful evaluation method is to construct a fictional guest journey in enough detail that every step can be observed and audited. The following scenario is illustrative only; it is proposed as a testing framework rather than a description of how any particular system behaves.

Suppose a caller contacts the restaurant on a Thursday morning asking for a table for four on Saturday evening at seven-thirty. The booking agent, in this scenario, should check availability against the seating inventory, confirm a specific table or table type, state any time limitation clearly, and deliver a confirmation reference that appears in the host's reservation view within a defined period. Each of those four steps is observable. A reviewer can check whether the confirmation appears, whether it matches what the caller heard, and whether the table assigned is one the restaurant can actually deliver.

Now extend the scenario. The same caller rings back two hours later to say that two more guests will be joining, making the party six. The booking agent should, in principle, reassess availability for six at seven-thirty on Saturday. The reviewer should check three things: whether the original booking for four is cancelled or updated rather than left open alongside a new booking for six, whether the agent accurately identifies that the original table cannot accommodate six and proposes a genuine alternative, and whether the caller is told clearly when the change is confirmed versus when it is pending staff review.

A third call arrives: the caller wants to move the booking to eight o'clock instead. Again, the original slot should close and the new slot should open. After all three calls, the reservation system should show one booking, for six guests, at eight o'clock Saturday, with a clear status. Reviewing that record against what was said on each call is a concrete and repeatable acceptance exercise.

Party Size Changes and the Ownership Question

When a booking changes, responsibility for confirming the new arrangement needs to sit with a named role. This is an operational design question that any restaurant should resolve before deploying an AI booking agent, regardless of the technology involved.

Consider three different size changes and who should own each one. A booking for two increasing to four is likely within the agent's authority to confirm if availability supports it. A booking for six increasing to twelve moves into territory that may require a different room, a different menu arrangement, or a conversation with an events coordinator. A booking for three decreasing to one raises a different question: whether the table should be released to another party or whether a single guest should still occupy it.

Each of these cases needs a written policy before the booking agent goes live. The policy should state not only who confirms the change but what the caller hears if the change cannot be confirmed immediately. A message such as "your request has been noted and someone will confirm with you within a stated period" is more useful than a vague assurance, because it gives the caller a concrete expectation and gives the restaurant a measurable commitment to meet.

The policy should also address what happens when a caller makes a change request outside of staffed hours. If the restaurant is closed on Sunday morning but a caller tries to increase a Sunday lunch party, the agent should acknowledge the request, record it accurately, and ensure it reaches a staff member before the booking becomes relevant. A request that disappears into an unmonitored channel is operationally equivalent to no request at all.

Special Requests: Recording Them Without Confirming Them

Window seats, quieter areas, high chairs, wheelchair access, birthday arrangements and private room enquiries all share a characteristic: they may be possible, but the booking agent alone cannot promise them. The distinction between recording a request and confirming it is easy to state and surprisingly easy to blur in practice.

A useful approach is to define a vocabulary of request statuses that both the booking agent and the front-of-house team understand. A request can be received, under review, confirmed or declined. If the status is anything other than confirmed, the guest should not hear language suggesting it will definitely be provided. Phrases that soften without misleading include noting that the preference has been passed to the team and that the guest will hear back before their visit.

The reservation record seen by the host should show the request and its status separately from the core booking details. If a host looks at the record five minutes before a party arrives and sees only the headcount and time, a window table request recorded by the booking agent may have been lost. The record layout matters for operational reasons, not cosmetic ones.

Dietary requirements occupy a different position. A caller mentioning an allergy or intolerance is providing safety-relevant information. The booking agent's role is to record the detail accurately and route it to a staff member who can respond. The decision about what the kitchen can safely prepare belongs to trained staff, not to an automated agent, and the evaluation of any booking system should assess how clearly that handover happens rather than whether the agent can answer food preparation questions.

Locations, Branches and the Risk of Ambiguity

Restaurant groups operating more than one site face a specific risk with AI booking agents: a caller may not be precise about which location they intend, and the agent may confirm availability at the wrong one. This is worth treating as its own evaluation category rather than a subset of general accuracy.

A practical test involves calling with a location reference that could apply to more than one site, for example a suburb name shared by two branches or a colloquial name used for the group as a whole. The useful outcome is that the agent asks a clarifying question and records the answer explicitly in the reservation. A confirmation that omits the specific branch name leaves the guest uncertain and the host unable to plan.

The reservation record field for location should therefore display the full and unambiguous site name, not a shorthand or a system code. When a host prints or reviews a booking sheet, a reference that requires cross-checking against another system introduces unnecessary steps during a busy service. A record that reads as a complete sentence of information is more reliable than one that requires interpretation.

Multi-location businesses should also consider how calls are routed before they reach the booking agent. If a caller rings a central number and the agent does not know which site the caller is associated with, the first question the agent should ask is the location. If the caller rings a site-specific number, the agent should already have the location as context and should confirm it rather than ask. The evaluation of an AI booking agent for a restaurant group should test both routing paths.

Cancellations and the Problem of Ghost Bookings

A cancelled reservation that remains visible in the system as live is a practical problem with direct consequences. A host who sees a booking for eight at nine o'clock and holds that table through service, only to have no party arrive, has lost revenue from guests who could have been seated. The booking system's cancellation handling deserves explicit testing rather than being assumed to follow from confirmation handling.

A cancellation test should verify that the record shows a clear cancelled status, that the time slot returns to availability in the system, and that no follow-up confirmation is sent to a guest who has already cancelled. It should also test partial cancellations: a party of eight reducing to five is not the same as a cancellation, but it should trigger the same reassessment of table assignment that a size increase would trigger.

Cancellations received after a certain point in the day may require a different response than those received days in advance. A restaurant with a late cancellation policy needs the booking agent to record the time of the cancellation request accurately, because that timestamp may be relevant to whether a fee applies or whether the table can be re-offered. The agent's role is to record and route; the decision about the policy application should rest with a staff member.

Preparing Front-of-House Staff for AI-Assisted Bookings

A booking agent that works well in testing may still create friction in service if the front-of-house team does not understand what it does and does not confirm. A brief, practical briefing for hosts and floor managers is worth preparing before the agent goes live. The briefing should address what a confirmed booking looks like in the system, what a pending or unresolved request looks like, and how to handle a guest who arrives with a different expectation from what the record shows.

Staff should also know how to reach the booking record quickly during service. If the reservation view requires navigating several screens under pressure, the operational benefit of an accurate record is partially lost. The acceptance exercise before going live should include a timed walk-through by a front-of-house team member, not only by the person who configured the system.

Handover arrangements at shift changes are another practical consideration. A booking agent may handle calls across the full day, but the host reviewing those records at the start of an evening service may not have been present when the calls were taken. The record needs to carry enough context that a new reader can act on it without needing to replay a call or contact the guest again.

A short daily review of bookings taken by the agent, conducted by a manager before each service period, creates a structured opportunity to catch requests that need staff action, spot inconsistencies between what the caller heard and what the record shows, and release or reassign tables where cancellations or reductions have been recorded but not yet acted on. This review does not need to be lengthy; its value lies in regularity rather than depth. A front-of-house team that trusts the booking record because they check it every day will handle exceptions more calmly than one that treats every arrival as a potential surprise.

Related guides

Comparing business systems? Explore Sysevo scheduling or book a Sysevo demo against the same requirements. This is an independent Sysevo editorial guide based on the linked vendor documentation, reviewed 2 November 2026. Examples are illustrative; no hands-on vendor test or guaranteed result is claimed. Sysevo publishes this guide and sells its own software.

For current Sysevo subscriptions and usage options, compare plans and pricing.