How Zogu answers your phone and takes the order

01What happens on a call

Phone line → speech recognition → Zogu checks your live POS menu → spoken reply → your POS

1. A customer calls your restaurant's Zogu number.
2. Speech recognition turns their words into text and decides when they've finished.
3. Zogu looks up what they asked for in your live POS menu.
4. A voice speaks the reply, ending with a clear next step.
5. After one read-back and a clear yes, the order goes into your POS.

Each restaurant has its own number, so the caller gets that restaurant's menu, hours and greeting: "Thanks for calling", your name, then "this is the automated ordering assistant." A number that isn't set up hears "Sorry, phone ordering isn't available right now. Please call back later. Goodbye." Zogu never guesses.

02How Zogu hears the caller and checks your menu

The hard part on a phone line is knowing when someone has finished.

• Zogu waits through a pause of a few seconds. "Yeah" or "mhm" doesn't stop it; "actually, no garlic bread" does.
• Zogu's own code searches your menu for the caller's words and offers only what can be ordered right now. The AI model never does the sums.
• "Twelve inch margherita" finds MARGHERITA 12", and shorthand is said as words: "three-piece Baklava".
• When the menu has no answer, it says so: "Sorry, I don't have the ingredients for the Mega Burger. Would you like to order one?"
• While it looks something up, the caller hears quiet typing, so a gap doesn't sound like a dead line.

Calls come in over Twilio. Speech recognition (Deepgram Flux), the AI model (GLM-4.7-Flash) and the voice (Deepgram Aura-2) run on Cloudflare Workers AI. Each sentence is checked before it's spoken, so working notes, tool names and brackets are never read out. More on how Zogu works from your live POS menu and how Zogu understands callers.

03The read-back and the order in your POS

Once the caller says that's everything, Zogu can make up to two add-on offers (how add-on offers work). Then one read-back, from our 1:35am test call (the whole call is summed up on the home page; "pickup" stands in for the word it used):

It gives the total and how it's paid, then asks "Shall I place the order?" A clear yes places it. Correcting the name places it without a second read-back.

• The order reaches your POS with items, choices, notes, pickup or delivery, and the caller's name and number, marked to be paid on pickup or delivery.
• Zogu gives the total and ready time, says goodbye and hangs up. It doesn't wait for the kitchen.
• One known bug: "it's Sam, and add a coke" places the order before the coke goes on.

More on how the read-back works, or try it on the demo page.

04How long callers wait for a reply

123 replies across 14 test calls, 8:27pm on 4 November to 1:35am on 5 November 2026, timed from the caller stopping to Zogu's first sound leaving our side.

The slow end is mostly the AI model. The test system had each of the seven test orders 0.64 to 0.94 seconds after Zogu started placing it. Our team made the calls, several builds were live, and the phone network adds time we can't see.

05When something goes wrong

Whatever breaks, the caller hears something. A slow AI model never ends a call on its own.

One case isn't settled: if the POS doesn't reply in time, the order may still have gone through, so we mark it unknown for a manual check. None of these lines offers a transfer or a call-back (what Zogu doesn't do).

06Setting up your restaurant

Five steps, all tried on our integration platform's test system except forwarding a line. No real restaurant has been through them.

1. Connect Zogu to your POS. Menu changes there reach the next call.
2. We check your menu in our console and add delivery areas if you deliver (pickup and delivery by phone).
3. We give you a number, greeting and voice (any of Deepgram Aura-2's 40 English voices).
4. You forward your line to it. We haven't done this for a real restaurant yet.
5. You test it and we fix what it missed on the Teach page.

The Teach page links phrases to items ("cheeseburger" means your Mega Burger) and sets how dish names are said, from the next call (the Teach page and call reviews). The console is internal for now.

07What you see after each call

The order arrives in your till, and you accept or reject it as usual. Our console keeps each call as text: what was said, reply times, where the caller cut in, the order as sent and how it ended. No audio is recorded. More in the FAQ.