If you run Square, the question that decides whether a voice agent earns its keep is narrow: does the order arrive in Square correctly, priced the way your register prices it? Everything else, the voice quality especially, is downstream of that. A ticket with a dropped modifier costs you food, a remake, and a customer, which is strictly worse than a call that went unanswered.
Here is how the connection works, where it tends to go wrong, and the specific things to test.
The connection is direct, and that matters
X1 Voice talks to Square directly rather than through a middleware layer. It pulls your item library, including variations and modifier sets, holds the conversation against that data, and writes the completed order back into Square as an order object your staff sees exactly like any other.
Directness buys you two things. Fewer moving parts between the caller and the ticket, and fewer places for a menu change to get stuck in transit. It also means the price the caller hears is the price in your item library, not a copy of it that someone updated last March.
That round trip is the actual product. A voice model that sounds warm but hands your staff a transcript to key in has moved work rather than removed it, which is the point made at length in why POS integration depth matters more than voice quality.
Where the ticket goes once it lands
Worth deciding before you go live, because a correct order in the wrong place is still a problem.
If you run Square KDS, phone orders appear as tickets on the display with the rest of your volume, and you can choose whether they land in the same queue as counter orders or their own. If you print, they print. Either way the question to answer with your kitchen manager is timing, not routing: a phone order placed at 6:15 for a 6:45 pickup should not fire the moment it arrives during a rush. Square KDS and phone orders goes through how operators usually set that up.
What a busy Saturday actually looks like
The change is not silence. It is the end of triage.
Picture 6:30 on a Saturday. Two people at the counter, one card reader thinking about it, and the phone going. Today somebody picks up mid-transaction, says "can you hold," and loses either the caller or the guest in front of them. Two more calls come in while that is happening and hear ringing.
With the agent answering, all three calls get picked up on the first or second ring. The two orders write into Square with pickup times. The third caller, who wanted to know whether you have a gluten-free option, gets whatever answer you configured, and if you never configured one, that call escalates to the counter, which is one interruption instead of three.
Settle the handoff policy before go-live, because it is the part operators skip and then regret. Complaints, large catering inquiries, and anyone who asks twice for a person should reach a person, and the agent should give up gracefully rather than looping through one more clarification. Write those rules down, test them by calling in angry, and check that the escalation actually rings somewhere a human will hear it during a rush.
Do the arithmetic on your own shift. Calls per shift multiplied by your real average handle time is the number of counter minutes you get back, and it is usually smaller and more useful than the figure on a vendor slide.
The failure mode: menu drift
Everything matches on day one because you just imported it. Then the item library moves and the phone menu does not follow.
The three drifts that bite hardest are a seasonal item added in Square with a name no caller would say out loud, a price change that lands on the register before anyone thinks about the phone, and a sold-out item on a Saturday night that the agent happily keeps selling. None of these throw an error. They surface as an angry pickup or a reconciliation gap at close.
The fix is a rule, not a feature. Whoever owns the Square item library owns the phone menu, and any menu change gets a test call the same week. If you sell fast-moving daily items, read real-time 86ing and menu sync and set the phone menu conservatively until you have measured your own sync lag.
Payment on the call
Callers can pay by card while they are on the phone for pickup and delivery, and the payment reconciles with your other Square tender types, so close does not grow a new manual step.
The part operators underrate: nobody on your staff ever handles a card number. No digits repeated across a counter, no numbers written on a ticket that sits by the register until close. That whole exposure path disappears, and collecting payment over the phone safely covers what changes in practice.
The four numbers to watch in month one
Answered-call rate moves first and is the reason you bought this. Order accuracy matters most, and you measure it by pulling twenty phone tickets and checking them against the transcripts rather than trusting a summary screen. Escalation reasons are your work list, sorted into handoffs that were correct by design and handoffs that were a defect you can fix. Average phone ticket tells you whether the agent's prompts do anything, and it is the number most easily flattered by a small sample.
One comparison worth making early, since you already run Square: how phone orders behave against your online ones. They usually differ in ticket size and item mix, and Square Online versus phone orders explains why treating them as one channel hides the interesting part.
Setup, honestly
Authorizing the connection takes minutes. Being ready to go live takes a day.
You connect Square, import the item library, set hours and any phone-only rules, run test calls including one deliberately awkward order, then forward your number. Most single locations are live inside 24 hours, and our team will do the menu pass if you would rather not. Plans start at $250 per month, and the Square integration page has the current detail.
The five-minute test
Before you sign with anyone, call the demo line and place the worst order your kitchen sees. No onions, sub fries, add a drink, apply the promo, change the size halfway through. Then look at what came out in Square.
If the ticket is right, everything after that is configuration you can tune. If the voice read your order back beautifully and the ticket is wrong, no onboarding call will fix it, and you found out in five minutes rather than after a season of remakes.