At peak times, the phone at a restaurant that delivers is engaged just when the orders would be coming in one after another. An AI phone ordering system answers several calls at once, takes the order from your menu and sends it to the kitchen in the same form as one placed on your website.
The caller learns from the first sentence that they are talking to a program. The assistant works only with what is on the menu, reads the order back before sending it and hands the call to a person when it does not understand or when someone mentions an allergy. It does not replace the person at the till: it takes the simple orders off their hands.
One line and one person, on a Friday evening
Phone orders all come at once: when it rains, on match nights, on Fridays at dinner time. An engaged line leaves no trace, so you do not know how many people called and gave up. And those who order by phone are often people from the neighbourhood, for whom you pay no commission to a platform.
The second problem is haste. An order taken standing up, with kitchen noise in the background, easily goes out with an incomplete address. The mistake only shows at the customer’s door, when it costs an extra trip.
The path of an order, from “hello” to the ticket in the kitchen
The caller can speak freely: “two large pizzas, one without mushrooms, and a cola”. The assistant matches what it hears to the items on the menu and asks only for what is missing: the size, the base, the sauce.
The most important part is the read-back. Before sending anything, the assistant repeats the items, the address and the total calculated by your ordering system, then waits for a clear “yes”.
- The items: name, size, options and extras, only from those on the menu and available at that moment.
- The address: street, number, block, staircase, floor, intercom, with a check that it falls within your delivery area.
- The phone number: the number the call comes from is suggested automatically, and the caller confirms it or gives another.
- Payment: cash or card on delivery; card details are never dictated over the phone.
- Sending: the confirmed order goes into the kitchen’s system and, on request, a text message goes out too.
The menu and the delivery area, in a form a program can use
The assistant needs the menu in structured form: each item with its variants, its compulsory and optional choices, and the price from your system. We also add the names people actually use: “a diavola”, “the one with tuna”, “the daily menu”. Without these synonyms, speech recognition catches the words but the program cannot find the item.
The link to the kitchen runs through an API, the bridge by which the assistant talks to the system where you already receive orders: your till software, your website’s ordering platform or the kitchen display screen. If your system cannot accept orders from outside, the fallback is the ticket printer or a tablet, to be checked with the software supplier.
Street names are where speech recognition goes wrong most often. That is why the assistant compares the address with the list of streets in your delivery area and says it back.
Allergies, special requirements and the moment a person takes over
There is one rule we do not leave to the program’s discretion: any mention of allergies, intolerances or a diet prescribed by a doctor stops the automated order. The assistant does not say that a dish “does not contain” something. It transfers the call or flags the order, so that the customer is called by someone at the restaurant before the food is prepared.
The same applies to requests that are not on the menu, to complaints and to very large orders. How allergen information is communicated over the phone is worth checking with your food safety specialist.
A transfer also happens when the assistant simply does not understand: noise on the line, someone who changes the order several times. After two failed attempts, the call goes to a person.
AI phone ordering for restaurants: the gradual start and what to track
We do not move every call to the assistant on the first day. We start with the ones that find the line engaged, on busy evenings, and follow the orders one by one.
If you want to keep recordings, the greeting says so plainly. The name, phone number and address are personal data: you keep them for as long as you need them for the order and for your records, under the terms of GDPR.
- Before going live: the structured menu, the delivery area and the transfer rules, written together.
- Tests: your team orders by phone the way people really do, with abbreviations, haste and changes of mind.
- The first period: only calls that are waiting, with a person checking every order that comes in.
- What you track: orders sent without corrections, transferred calls and the reason for each, abandoned orders.
Frequently asked questions
What happens if the assistant gets the order wrong?
That is why it reads everything back and waits for confirmation before sending. Mistakes can still happen, as with any order taken over the phone, and the initial period is for finding them.
Can it work with the till software I use?
It depends on whether the software can accept orders from other applications. Some have such an interface, others do not, so we check with your supplier before anything else.
What does the monthly cost of an order-taking assistant depend on?
On the number and length of calls, because telephony, speech recognition and speech synthesis are charged by usage. At the start, there is also the work of structuring the menu and connecting to the ordering system.
This is a typical project description: it shows how we usually approach this kind of work and does not present a project carried out for a particular client. Every real project starts from your company’s situation, and the stages, timescales and price are agreed after the initial discussion.