The phone that rings during dinner service
Picture a restaurant at half past six in the evening. Full house, the kitchen flat out, and the phone is ringing. Someone wants a table for four on Saturday. The waitress rushes past with three plates and hasn't got a spare hand. The phone rings out. The booking — and maybe the customer — is gone.
This is exactly the situation a voice AI assistant is built for. Not as a replacement for a person, but as someone who picks up precisely when people can't keep up. It takes the booking, writes it into the system, answers a question about opening hours. And when the matter is more complex, it hands it on to a human.
What it handles surprisingly well today
The technology has matured a lot over the past two years. Today's voice assistant understands fluent speech, doesn't talk over you and sounds natural — a long way from the robotic voices of five years ago. Short, clearly bounded tasks it handles reliably.
Typically that means bookings (a table, a hairdresser slot, a car inspection), simple orders, and answers to repeated questions — where are we, how late are we open, do you have step-free access. In these scenarios it can clear a large share of calls without the caller feeling fobbed off. Crucially, it's never engaged and never tires on the twentieth identical query in an hour.
Where it still grinds
Let's be honest, because overblown promises will hurt you more than help. The assistant struggles when a call drifts off-script. A strong accent, a dialect, a call from a noisy street, or an agitated customer talking a mile a minute — all of these lower recognition success.
It also struggles with complex, multi-layered requests: "I want to move my booking from Friday to Saturday, but only if you've got the same table by the window, otherwise cancel it." A human gets it; the assistant may stall. And there's sensitivity: with a complaint or an emotional situation, the customer doesn't want a robot, they want a person. A good system recognises this and hands the call over — it doesn't try, at all costs, to see through something it isn't equipped for.
Handover to a human is a feature, not a failure
The most important part of a good voice assistant isn't what it resolves on its own, but how cleanly it hands a call over when the moment comes. The customer shouldn't have to repeat everything from scratch. The assistant should pass on the context: who's calling, what they want, what's already been said.
Set clear rules for when a call routes to a person — on a complaint, on a repeated misunderstanding, on an explicit "I want to speak to a human." When you're closed for the morning, the assistant takes a message and sends it to email or into the CRM. The goal isn't for the AI to handle a hundred percent. The goal is that not a single call goes unanswered — it handles part itself and passes the rest on cleanly.
When it makes sense, and when it doesn't
A voice assistant pays off where there are many short, repeated calls and where a phone ringing out genuinely costs you money — restaurants, clinics, repair shops, anywhere with bookings. If ten calls a day ring out because nobody can pick up, each one is a possible customer.
Conversely, if you take a handful of calls a day and each is different, complex and personal — consultancy, bespoke individual sales — the investment won't pay back and you risk coming across as cold. Here's the honest answer: don't do it. Automation should serve where volume and repetition make sense, not everywhere it's technically possible.
How to approach it soberly
Start small. Don't try to cover every call at once. Pick one narrow scenario — say, just bookings after closing time — and let the assistant do that one thing well. Watch the records for a month: how many calls it handled itself, how many it passed on, where it got stuck.
Those records are how you tune the system. You add further scenarios gradually, as trust in the numbers grows. This cautious approach is duller than "deploy AI and done," but it's the difference between a tool that helps your customers and one that drives them away. In practice this narrow, well-measured step is usually where we begin.