If you run Toast, one question decides whether a voice agent is worth anything to you: does the order land in Toast correctly? A pleasant voice that drops a modifier and sends the line the wrong ticket is worse than a call that rang out, because now you have paid for food and a remake instead of nothing.
Here is what the connection actually does, where it breaks, and what to test before you sign.
What the connection actually does
X1 Voice reaches Toast through Deliverect, the same middleware layer a lot of restaurants already use to pull third-party orders into their POS. The agent pulls your live Toast menu, including modifier groups and prices, holds the conversation against that menu, and writes the finished order back as a ticket.
The round trip is the whole point. A voice model that sounds human but hands you a text message with an order in it has not saved anyone any work, because someone still has to key it in during a rush. The value sits in reading the live menu and firing a correctly-modified ticket, which is a data problem, not a speech problem.
Toast menus tend to be deep. Nested modifier groups, required choices, forced sizes, upsell prompts attached to specific items. That depth is good for the register and hard for a voice conversation, because a caller does not know your modifier tree exists. Before you go live, it is worth reading through Toast modifier mapping for voice and deciding which branches a caller should ever be asked about out loud.
What a Friday night looks like with it on
The change is not that the phone stops ringing. It is that the phone stops interrupting.
At 6:40 on a Friday, your host is seating a four-top with a line behind them. Three calls come in over about ninety seconds. Two are pickup orders, one is someone asking whether you are still doing the fish special. Without an agent, the host answers one of those, puts a guest on hold to do it, and the other two callers hear ringing and try the place down the street.
With the agent running, all three get answered on the first or second ring. The two orders land in Toast as tickets with pickup times. The special question gets an answer from whatever you configured, and if you did not configure one, that call escalates to the host, who now has one interruption instead of three.
One thing to decide before go-live is who the agent hands off to and when. Complaints, large catering inquiries, and regulars who want a person should reach a person, and that is a policy you set rather than a limitation you discover. Human handoff and failover covers how to write those rules so the agent does not argue with a caller who has asked twice for a manager.
The staffing effect is real but modest and worth doing honestly. Roughly, take your phone calls per shift, multiply by average handle time, and you get minutes returned to the floor. Do that arithmetic with your own numbers rather than a vendor's, and use your real average handle time, not the one you wish you had.
The failure mode nobody warns you about
Menu drift.
The connection works on day one because you just imported the menu and everything matched. Then six weeks pass. The kitchen 86s the short rib every Saturday by 8pm. Marketing adds a limited-time sandwich in Toast but nobody checks whether the agent can hear the word for it. A price goes up two dollars on the register and the phone quote stays where it was.
None of those show up as an error. They show up as a caller arriving to pick up something the kitchen cannot make, or a fifteen-cent tax discrepancy at close that nobody wants to chase. The fix is procedural, not technical. Whoever owns the Toast menu owns the phone menu, and menu changes get a test call the same week. The specific ways sync goes wrong are catalogued in POS 86 sync failure modes, and it is short reading before you commit.
Payment on the call
Callers can pay by card during the call for pickup and delivery, and the payment reconciles with your other tender types, so end-of-day close does not sprout a new pile to reconcile by hand.
The quieter benefit is that nobody on your staff ever hears a card number. The classic restaurant exposure is a host repeating digits back across a counter, or writing them on a ticket that sits by the register until close. That path disappears. If your processor or franchisor asks about scope, collecting payment over the phone safely covers what actually changes.
The numbers to check in the first month
Four, in this order.
Answered-call rate is the first, because it is the only number that moves immediately and it is the reason you bought this. Order accuracy is the one that matters most, and you measure it by pulling twenty phone tickets and checking them against transcripts, not by trusting a dashboard. Escalation reasons tell you what to fix, sorted into handoffs that were correct by design and handoffs that were a defect. Average phone ticket tells you whether the agent's prompts are doing anything, and it is the number most likely to be flattered by a small sample.
If accuracy is holding above your own pre-agent baseline and answered-call rate went from something like 70 percent to near 100, the system is working. If accuracy slipped, stop and fix the menu before you look at anything else.
Setup, realistically
The mechanical half of this — how the menu is read out of Toast, how the ticket is written back, and which modifier structures cause trouble — is covered in Toast menu sync and modifier mapping. Read it before your configuration call if you maintain the menu yourself.
Connecting the account is minutes. Being ready to go live is a day.
You link Toast through Deliverect, import the menu, set hours and any phone-only rules, run test calls including one deliberately awkward order, and then forward your line. Most single locations are live within 24 hours. If you would rather not do the menu pass yourself, our team handles it.
A mixed estate is not a blocker. X1 Voice connects directly to Square, Clover and OrderCounter, and reaches Toast, Lightspeed, TouchBistro, SpotOn, Aloha, Revel, PAR Brink and Micros through Deliverect, so locations on different systems can run the same phone experience. The full list is on the integrations page, and plans start at $250 per month.
The test that decides it
Before you buy anything, from us or anyone, call the demo line and place the worst order your kitchen sees. No onions, sub fries, add a drink, apply the promo, and change your mind about the size halfway through. Then look at the ticket that came out the other end.
If the ticket is right, the rest is configuration. If the voice repeated your order back perfectly and the ticket is wrong, no amount of setup help will fix that, and you have learned it in five minutes instead of five months.