Short answer
AI reservation enquiry handling is a workflow where a model reads booking and function enquiries from email, web forms and social messages, extracts the date, party size, occasion and dietary needs, checks availability in your booking system, and drafts a reply in your venue’s voice. Staff approve group and function replies, take deposits through the booking platform, and the kitchen, not the model, answers every allergy question.
- Standard bookings can be answered automatically; groups and functions are drafted for staff.
- Allergy and dietary notes go verbatim to the kitchen; the model never promises an item is safe.
- Card details never travel by email or chat; deposits use the booking platform’s payment link.
- A booking confirmation is not consent to marketing under the Unsolicited Electronic Messages Act.
What the model handles and what it leaves
AI reservation enquiry handling covers the messages your online booking widget cannot: “Can we do 14 people on the Saturday before Christmas with a set menu, two vegans and one coeliac?” arriving by email at 9pm, or an Instagram message asking to move a booking. The model reads the message, pulls out the facts, checks availability in the booking system and drafts a reply. What it may send on its own depends on the lane.
| Enquiry | Lane | Notes |
|---|---|---|
| Standard table, within normal hours and party size | Reply with availability and booking link | Booking made through the platform, not by the model |
| Change or cancellation within policy | Confirm automatically, log in the platform | Outside policy goes to staff |
| Group of 7 to 20 | Draft for manager approval | Set menu, deposit, seating |
| Function or exclusive use | Draft for events manager | Minimum spend, contract, run sheet |
| Allergy or dietary need | Flag to kitchen, reply that the team will confirm | Guest’s words copied exactly |
| Complaint about a past visit | Manager only | No automated reply beyond acknowledgement |
How the workflow runs
- Inputs
- Emails, website forms, Facebook and Instagram messages, voicemail transcripts, current menus and set-menu options, booking and deposit policy, and availability from the booking system.
- Systems
- A booking platform such as Now Book It, ResDiary or SevenRooms (read availability; create bookings only through its own flows), the shared inbox, a review queue on a phone, and a model API.
- Steps
- Classify the enquiry; extract date, time, party size, occasion and dietary notes; resolve relative dates (“next Friday”) against the calendar; check availability; draft a reply with options; route by lane; log the outcome against the booking.
- Human review
- The duty manager approves group drafts, the events manager owns functions, and the head chef or kitchen lead answers every allergy question before the booking is confirmed.
- Outputs
- Sent replies for standard enquiries, approved drafts, bookings with dietary notes attached, and a daily list of open enquiries.
- Measures
- Median time to reply, group enquiries converted to bookings, double bookings, and corrections made to drafts.
- Risks
- Wrong dates, double bookings, invented menu or allergen claims, card details in email, and marketing sent without consent.
NZ rules that shape the replies
Allergens come first. Food Standards Australia New Zealand’s allergen labelling guidance says that for food not required to bear a label, such as food at a café or takeaway, allergen declarations must be displayed with the food or provided to the purchaser on request using the required names. Your kitchen’s current recipes are the only valid source. The model copies the guest’s words into the booking, flags it, and says the team will confirm.
Dietary and health details (a coeliac diagnosis, a pregnancy) are personal information under the Privacy Act 2020. Collect only what the kitchen needs, tell guests why, and do not keep it in a marketing database. A follow-up offer is a commercial electronic message under the Unsolicited Electronic Messages Act 2007, which bars unsolicited ones and puts the burden of proving consent on the sender.
Illustrative example: a 90-cover Wellington restaurant
Illustrative example, synthetic numbers: a restaurant with a 24-seat private room receives 35 group and function enquiries and 120 standard booking messages a week. Assumptions: a manager spends 9 minutes on each group reply and 3 minutes on each standard one; approving a draft takes 3 minutes for groups and 1 minute for standard messages.
- Before: 315 + 360 minutes = about 11 hours a week.
- After: 105 + 120 minutes = under 4 hours a week, so roughly 30 hours a month back.
- Model fees: about 670 messages a month at 3,500 input and 500 output tokens on Claude Haiku 5.5, at US$0.10 and US$0.50 per million tokens (Anthropic, as listed on 19 October 2026), is under US$1 a month.
The bigger gain is speed. Drafts are ready within minutes, so a manager can approve them between services instead of the next morning.
What stays with a person
Staff set and approve every group and function quote, minimum spend, deposit and contract. The kitchen answers allergy questions. Managers handle complaints, refunds of deposits and any exception to policy. The model never takes payment details and never creates a booking outside the platform’s own process.
Where enquiry handling goes wrong
- Date confusion. “05/06” means 5 June here and May 6 to an American visitor. Confirm dates in words.
- Double bookings. If the model holds tables in a spreadsheet while staff book in the platform, they will collide. One source of truth only.
- Invented answers. Asked about corkage or a menu item, the model guesses. Retrieve only from current, dated venue information.
- Payment details. Guests paste card numbers into emails. Redact them before processing and reply with the platform’s payment link.
The same pattern of inbox triage and drafted follow-up, with staff keeping decisions, runs in the property management agent we built for APL Property. For phone enquiries, compare a chatbot with a voice agent or see AI voice agents. Once guests arrive, a guest support knowledge assistant handles in-stay questions, and supplier invoice processing covers the back office. Our lesson on drafting emails with human review shows the approval habit, and the hospitality overview and AI customer support service cover the rest.
Sources
- Allergen labelling for food businesses, Food Standards Australia New Zealand, accessed 19 October 2026
- Unsolicited Electronic Messages Act 2007, New Zealand Legislation, accessed 19 October 2026
- Privacy Act 2020, New Zealand Legislation, accessed 19 October 2026
- Claude API pricing, Anthropic, accessed 19 October 2026
Written by the Levy Street team in Wellington. Last updated 19 October 2026.