AI for Cafés and Restaurants: Bookings, Reviews and the Phone

A café phone rings at ten past twelve. Every pair of hands is on a coffee machine, a till or a plate. The call goes unanswered, and so do the next four, and by two o'clock there are seven missed calls and no way to tell which ones were a table of eight on Friday and which were a supplier.

That is the shape of the problem in hospitality. It is not that you lack systems. It is that the busiest hour of your day is also the hour when the most enquiries arrive, and the people who could answer them are the people who cannot stop. AI for restaurants in NZ is mostly about that overlap, and the cheapest fixes are the ones worth doing first.

Margins in this industry do not tolerate a large technology budget. So this article is ordered by cost, lowest first, and I have been honest about the points where the spend stops making sense.

The phone during service is the single biggest leak#

Most venues I talk to have no idea how many calls they miss, because a missed call on a landline leaves no trace unless somebody goes looking. The first job is to find the number. A modern phone system will show it in a call log. If you run calls through a mobile, the log is already on the handset. Count a normal week and split the misses by hour.

The pattern is usually the same. Calls cluster between eleven and two, and again around five to seven, which are precisely the windows when nobody can answer. A booking for six on Saturday walks away to the next place on the search results, and you never learn it existed.

An AI phone agent answers those calls on the first ring, every time. It knows your hours, your address, whether you take walk-ins, what the parking situation is, whether the kitchen can handle a gluten-free table of ten on a Friday, and it can take the booking straight into your diary. Anything it cannot handle, a supplier dispute or a complaint, it passes to a human with the conversation attached rather than dumping the caller into voicemail. The mechanics of that transfer matter more than people expect, and I have written about getting the handoff right separately.

The economics are simple arithmetic you can do yourself. Take your missed calls for a week, assume a modest share of them were bookings, multiply by your average spend per head and the party size you typically get on the phone. Compare that to a few hundred dollars a month. In most venues the number is not close.

Bookings, changes and the no-show problem#

Taking a booking is the easy half. Managing what happens to it afterwards is where the money is.

A no-show on a Saturday night is a table you turned away someone else for. A late cancellation is nearly as bad. The fix is unglamorous: a confirmation message when the booking is made, then a reminder the day before with a one-tap way to cancel or change. Automating that sequence takes a booking platform and about an hour of setup, and it is the highest-return thing most venues can do this month.

Where AI adds to a standard booking platform is in handling the messy requests that a form cannot. Someone wants to move a table of twelve from seven to eight thirty. Someone wants to know whether the back room can be held for a birthday and whether they can bring a cake. Those arrive as calls and Instagram messages, and each one costs a staff member several minutes of attention during a shift.

There are real failure modes here, and pretending otherwise sets you up for an embarrassing Saturday. Large groups and anything touching a private area should require a human to confirm. So should any booking that would put you over a capacity you cannot actually seat. The guardrails that stop a booking agent embarrassing you are worth reading before you switch one on.

That last row is there deliberately. If a vendor is quoting you a five-figure integration for a single site, the numbers almost never work in hospitality. Get the cheap layers running first and let the gaps tell you whether more is needed.

Reviews are marketing, and you are already behind on them#

Google reviews do more for a local venue than most paid advertising, and the response rate on them is dismal because writing thoughtful replies is a job that never feels urgent.

AI is genuinely good at this specific task. Feed it the review, your venue's voice and a few rules about what you will and will not say, and it drafts a reply in seconds. You read it, adjust a line and post. A month of backlog clears in twenty minutes.

Two rules keep this from going wrong. Never let it post without a human reading the draft, because a generated apology to a serious complaint can read as dismissive and make things considerably worse. And never let it argue. A reply to a bad review is written for the hundred people who will read it later, not for the person who wrote it.

Responding consistently also feeds your visibility in local search, which sits alongside the basics covered in SEO for service businesses. A venue with recent reviews and recent replies looks alive. One with the last reply from 2023 looks like it might have closed.

Social messages belong in the same pipeline as calls#

Instagram is where a large share of hospitality enquiries now start. Someone sees a photo of a dish, sends a DM asking whether you are open Monday, and gets an answer eleven hours later when the owner next opens the app.

Speed matters here more than polish. The venue that replies in four minutes gets the booking. Routing DMs and comments into the same place your calls land, so one system answers them with the same information, removes the "who is checking Instagram today" problem entirely. There is more detail in the piece on answering comments and DMs without living in your phone.

Keep the tone recognisable. Hospitality is a personality business, and a message that reads like a bank's auto-reply undoes the reason someone messaged you in the first place. Write a short voice guide, feed it in, and read the first fifty replies before you trust it unattended.

Where AI for restaurants in NZ stops paying for itself#

A few things get pitched to venues that I would leave alone.

AI menu writing produces copy that reads like every other menu on the internet. Your dish descriptions are one of the few places your venue's voice reaches a customer directly, and generic adjectives cost you more than the twenty minutes they save.

Demand forecasting and AI-driven rostering are genuinely useful at scale. At one or two sites, your head chef's read on the weather and the school holidays is better than any model you can afford, and the setup effort is real.

Anything that automates a complaint response without a human in the loop should be off the table. Complaints are where you either keep a customer or create a public story, and neither outcome should be delegated.

The order I would do it in#

If you are starting from nothing, the sequence is roughly this. Get a booking platform sending automated confirmations and reminders, because it is cheap and it attacks no-shows immediately. Then work out what your missed calls are costing you and decide whether an AI phone agent clears that bar. Then clear the review backlog and set up a weekly habit for it.

Before signing anything, check what the vendor charges beyond the base plan. Voice minutes and setup fees are where hospitality quotes get away from people, and the pricing bands for AI chat and phone agents in New Zealand are worth reading first. It is also worth understanding how these tools fit the wider local picture in the guide to AI solutions for Auckland businesses.

The one thing to do this week costs nothing. Pull your call log, count the misses, and split them by hour. If the answer is uncomfortable, you already know which problem to solve first.