An AI phone answering service picks up your restaurant's calls in about one ring, understands what the caller wants in speech rather than menu presses, and either completes the booking or order itself or hands the call to a person. Skai Solutions runs that loop in 5 stages, and the stage that decides whether it is useful or infuriating is the third one, not the voice.
1. It answers. Your existing number diverts to the service. You keep the number and you control the divert, so it can be always on, during service only, or after hours only. Nothing is rewired and no handset changes.
2. It transcribes. The caller speaks and a speech model turns it into text in real time. This is where most systems built elsewhere fall over on Australian and New Zealand calls: suburb names, Māori place names, and the local habit of running words together defeat a model trained on North American English. A demo in a quiet room proves nothing about a Friday at 7pm with a coffee machine going.
3. It decides what the caller wants. This is the stage that matters. A booking, a takeaway order, opening hours, a dietary question, a supplier, or a person who needs a human. Get this wrong and the caller gets a confident, fluent, irrelevant answer, which is worse than voicemail because it wastes their time before they give up.
4. It writes the outcome somewhere real. A booking into your diary, an order into your POS. If it cannot write into a system you already run, you have bought a transcription service and someone still has to rekey everything. Skai Solutions connects to Kounta and Lightspeed, Square, Impos, H&L POS, Tevalis, Revel, TouchBistro and Oracle MICROS. That list is shorter than some, and if your system is not on it we say so rather than promise a date.
5. It transfers or logs. Anything outside its scope goes to a person, and every call leaves a transcript, a recording and an outcome you can audit. The audit trail is the part operators underrate. It is the only way to find out what the system is getting wrong before your customers tell you.
The strongest evidence on caller patience is not a vendor statistic. It is Brown, Gans, Mandelbaum and others, Journal of the American Statistical Association, 2005, a full-year census of more than 1,200,000 calls to a single call centre. About 20% of callers requesting service abandoned before being served, and mean caller patience was 446 seconds, roughly 7.4 minutes. The useful finding is the shape, not the average: hang-ups spiked twice, once in the first few seconds as the hold message began, and again at about 60 seconds when it repeated.
You will see a claim that 60% of callers hang up after 45 seconds on hold. Skai Solutions does not use it and neither should you. It traces to an unnamed, undated study cited by an on-hold audio vendor, and it is contradicted by the census above. It was on this site until August 2026 and it was removed for that reason.
For a sense of how many calls a hospitality business really drops, the closest measured figure is the Marchex Institute analysis of 5.8 million calls to 13 hotel brands: a 14% failed call rate, with 5% never answered at all. That is hotels, not restaurants, and the data is from 2017. It is the honest ceiling on what anyone can claim here, and it is why the sensible pitch for this category is answer rate rather than a dollar figure invented to sound alarming.
Four things, consistently, and any vendor who tells you otherwise has not run one through a real Saturday.
If your phone mostly carries supplier calls and general enquiries rather than bookings and orders, a basic message-taking plan is the right buy and it is not us. Message-taking plans advertised in this market start from about NZ$19 a month for a small allowance, advertised August 2026, and they do a simpler job well.
If you are a single-site venue taking under about 100 calls a month, the arithmetic rarely works: the fixed cost per useful call is too high. If you run a POS outside the eight listed above and the integration matters to you, wait. And if your calls are predominantly walk-in enquiries rather than revenue, fixing your online listings will return more than fixing your phone.
Two pieces of market context worth having before you spend anything. The CreditorWatch Business Risk Index measured a 10.4% closure rate across food and beverage services in the twelve months to January 2026, the highest of any industry and roughly double the economy-wide average. At the same time the ABS Counts of Australian Businesses recorded the number of Accommodation and Food Services businesses rising 1.3% in 2025-26. More venues, more failures, thinner margins. Fixed monthly costs are easier to defend in that market than variable ones, which is an argument for a plan with included minutes over per-call pricing, whoever you buy from.
| Approach | Caller experience | Where the outcome lands | Best for |
|---|---|---|---|
| Voicemail | Leaves a message, no confirmation | Nowhere until someone listens | Nothing, honestly |
| Press-one menus | Navigates a tree, cannot speak freely | A department, not an outcome | Multi-department businesses |
| Message-taking plan | Talks to a person, order not taken | An email or text to you | Enquiry-heavy phone lines |
| Human receptionist | Best experience available | Wherever you train them | Venues that can roster one |
| AI phone agent | Speaks freely, completes the task | Your POS or booking diary | Order and booking heavy lines |
Call abandonment: Brown, Gans, Mandelbaum and others, Journal of the American Statistical Association, 2005. Hotel failed call rate: Marchex Institute, 5.8 million calls to 13 hotel brands, 2017 data, published July 2018. Closure rate: CreditorWatch Business Risk Index, February 2026. Business counts: ABS Counts of Australian Businesses, 18 August 2026. Message-taking price is an advertised market rate observed in August 2026, quoted by category rather than by company.
Skai Solutions publishes its prices rather than gating them. The full plans, the GST treatment and the wage comparison are on the Australian pricing page and the New Zealand pricing page. If you want the category defined before the price, read what an AI receptionist is. If your volume is takeaway orders rather than bookings, the relevant build is phone ordering into a POS, and the no-show side of the same problem sits on what restaurant no-shows cost.
Live in five days, month to month, 30 days' notice to cancel.
If you would rather keep the calls in house for now, our guide to answering the phone at a restaurant sets out the greeting, the six fields and the four fixes in order of cost.
Your number diverts to the service. It answers in about one ring, transcribes what the caller says, works out whether it is a booking, an order, a question or a person who needs a human, writes the outcome into your POS or booking diary, and transfers anything outside its scope. Every call leaves a transcript and a recording.
That depends entirely on what the speech model was trained on. Systems built for North America commonly mishandle suburb names and Maori place names on real calls even when they demo well in a quiet room. Test it on your own line with a real order before you sign anything.
It should transfer to a person rather than guess. A confident wrong answer is worse than voicemail because it wastes the caller's time before they give up. Test the transfer path during a demo, not after go-live.
No. You keep your existing number and divert calls to the service. You control when the divert is active, so it can run always, during service only, or after hours only. Skai Solutions takes five days to set up.
Skai Solutions connects to Kounta and Lightspeed, Square, Impos, H&L POS, Tevalis, Revel, TouchBistro and Oracle MICROS. If a service cannot write into a system you already run, someone on your team still has to rekey every order.
No, and it should not be. Message-taking plans in this market advertise from about NZ$19 a month for a small allowance as at August 2026, because they take a message rather than complete an order. If your phone mostly carries enquiries, that is the right buy and it is not us.