Voice AI · missed-call math

The math on voice AI for restaurant missed calls, with the actual formula

Every pitch quotes a six-figure annual leak from missed calls. None of them show their work. This is the formula, the inputs you can pull yourself, the sensitivity tables for shops of different sizes, and the break-even arithmetic against a real $350-a-month price tag. Plug in your numbers, decide for yourself.

Matthew Diakonov, Written with AI

Published May 21, 20265 min read

Direct answer (verified 2026-05-21)

How do you calculate voice AI ROI on restaurant missed calls?

Weekly revenue leak equals weekly inbound calls times miss rate times order-intent share (~0.6) times average ticket. Voice AI pays for itself when recovered orders times ticket exceeds the monthly fee. At PieLine’s $350-a-month flat rate, that is 10 recovered orders at a $35 ticket, which is a 1 percent recovery on a 1,000-call month. Anything above that line is margin.

The formula, four inputs, one number

You only need four numbers. Two come from your VoIP provider, one comes from your POS, one is an industry constant you can calibrate later. Anybody trying to sell you a missed-call solution should be able to walk you through this on a whiteboard. If they cannot, that is a signal.

missed-call math, paste in your numbers

From your VoIP dashboard

calls_per_week: total inbound
miss_rate: unanswered / total

From your POS and an industry average

avg_ticket: takeout / delivery average
order_intent_share: ~0.60, refine later

Sensitivity: what the leak looks like for your size of shop

The same formula across three common operator shapes. Order-intent share held constant at 60 percent. Miss rate at 30 percent for the small shop, 35 percent for the mid, 40 percent for the high-volume pizza/Indian/Chinese shop where every Friday rush rolls to a busy signal.

Shop profile	Calls/wk	Miss %	Ticket	Weekly leak	Annual leak
Small independent (one line)	100	30%	$28	$504	$26,200
Mid-volume takeout / Chinese / Mexican	250	35%	$35	$1,838	$95,500
High-volume pizza / Indian / Italian	500	40%	$45	$5,400	$280,300
Multi-line / multi-location, peak only	800	25%	$55	$6,600	$342,800

Annual leak uses 52 weeks. None of these numbers are aggressive. Drop the miss rate to 20 percent on the worst row and you still leak $171K a year. The formula is robust to halving any single input.

Break-even against a real $350-a-month price tag

The other half of the math nobody shows you. The leak is meaningless if the cure costs more than the leak. PieLine charges $350 a month flat for the first 1,000 calls, then $0.50 per call beyond that. That is a fixed number you can divide into, which is rare in this space, most competitors price per minute. The break-even math is one division.

breakeven.ts

0 orders

recovered per month at a $35 ticket covers the entire $350-a-month fee. On a 1,000-call month, that is a 1 percent recovery rate.

That is the part that turns the whole exercise from a marketing calculation into an obvious one. The disagreement is never about whether the fee is worth it. The disagreement is about whether the recovery rate is real, which is the question production accuracy actually decides.

The math without voice AI vs the math with it

Same shop, same week, same calls. The only thing that changes is whether anyone picks up. Use the toggle to flip between the two scenarios at a 250-call-per-week mid-volume shop.

Same shop, two phone realities

Friday rush is a single host juggling counter, takeout, and the phone. The phone loses every prioritization fight. Voicemail rolls. Customers hang up.

250 calls/wk, 35% miss = 87 missed
60% order-intent = 52 lost orders
$35 ticket = $1,830/wk lost = $95K/yr
85% of missed callers never call back
Negative reviews: 'called 3 times, nobody picked up'

How to verify these numbers on your own line in one week

Do not take any of this on faith. The whole point of putting the formula on the page is that you can falsify it inside a week, cheap. Three steps.

Pull seven days of call analytics from your VoIP provider (Nextiva, RingCentral, OpenPhone, Google Voice, Vonage). Filter by 11am to 1pm and 5pm to 8pm. Count total calls and unanswered calls. If you are still on a landline with no analytics, run a tally sheet by the phone for seven days, that gets you a usable number too.
Pull your average ticket for takeout and delivery from your POS over the same week. Do not use dine-in tickets, they bias the number high. Average just the phone-relevant ones.
Multiply and decide. Calls times miss rate times 0.60 times ticket. That is your weekly leak. Compare to $350. If the weekly leak is more than the monthly fee, the math is over.

The pieces of this that are not algebra (recovery rate, accuracy, POS fit, what happens during a complaint call) are real and worth interrogating. We’ve written separately about where the industry leak numbers come from and why text-back and voicemail rescue do not move the needle.

What the math actually looks like at 11 locations

Mylapore is an 11-location South Indian chain in the Bay Area rolling PieLine out across every store. The owner has publicly endorsed the rollout and reports projecting about $500 of additional revenue per location per day from eliminating the phone bottleneck. Multiply through: $500 times 11 stores times 365 days is roughly $2 million a year of recovered revenue against a per-store voice AI fee that is, in P&L terms, a rounding error. The same arithmetic is what every operator ends up running once they take an hour to pull the inputs.

The reason the per-store math works that hard at Mylapore is the same reason it works for any concept where the phone is the bottleneck during rush, not the constraint on the rest of the day: the AI sits in for the staff hour that costs the most labor and the staff attention that costs the most missed orders. The number you should care about is not the headline, it is the one you produced from your own VoIP dump.

Run the numbers on your own line

Bring a week of VoIP analytics and your average takeout ticket. We will plug your inputs into this formula on the call and show you exactly what the break-even looks like on your menu.

Frequently asked questions

What is the simplest formula for restaurant missed call revenue lost?

Weekly revenue lost = calls per week times miss rate times order-intent share times average ticket. The order-intent share is the fraction of inbound calls that are actually trying to place an order or reservation, not asking about hours or directions. Industry data puts it around 60 percent. A 250-call week at a 30 percent miss rate with a $35 ticket leaks $1,575 per week, which is $81,840 per year. Plug your own numbers in. Every input is something you can pull from a VoIP dashboard or your POS in an afternoon.

How many orders does voice AI have to recover for it to be worth it?

Divide the monthly fee by your average ticket. PieLine charges $350 a month flat for the first 1,000 calls. At a $35 ticket, break-even is 10 recovered orders per month. At a $25 pizza ticket it is 14. At a $55 Indian or Italian ticket it is 7. For a shop doing 1,000 calls a month, 10 recovered orders is a 1 percent recovery rate on inbound, which is well below any documented baseline miss rate.

Why use 60 percent as the order-intent share instead of 100 percent?

Not every call is an order. Restaurants get callers asking about hours, directions, dietary questions, takeout vs dine-in confusion, asking to speak to a manager, wrong-number, and so on. The published share of calls that are actually trying to transact is around 60 percent for takeout-heavy concepts and higher for delivery-heavy pizza shops. If you use 100 percent you will overestimate the leak. Sixty percent is the conservative number every operator should anchor on first.

What miss rate should I use if I do not have VoIP analytics?

Use 30 percent as a baseline and run a one-week tally sheet by the phone to verify. Industry studies put the average independent restaurant at 25 to 35 percent missed calls across all hours and 40 to 50 percent during the 5pm to 8pm dinner rush. Single-line shops sit at the higher end. Switching from a landline to a VoIP provider with analytics (Nextiva, RingCentral, OpenPhone, Google Voice) gets you the real number inside a week.

Does voice AI actually recover those calls, or does it just answer them?

Two different questions. Answering is the trivial part, a forwarding rule and a vendor that picks up. Recovery is the part that matters: ticket fires correctly, modifiers map to the real POS, price comes out right, the kitchen gets a usable order. PieLine targets 95 percent order-level accuracy in production, hands off the rest to a human with full context, and reads availability from a real menu so the agent cannot promise something the kitchen does not have.

What does Mylapore actually see in dollars from this math?

Mylapore is an 11-location South Indian chain in the Bay Area rolling out PieLine across all stores. The owner reports projecting around $500 of additional revenue per location per day from eliminating the phone bottleneck. Multiply 500 by 11 locations by 365 days and you get roughly $2 million a year, against a per-store voice AI fee that is a rounding error in any P&L. That ratio is the back-of-envelope reason any operator with phone-heavy traffic should run the math themselves.

What is the upper bound on the leak number?

Published estimates top out at roughly $292,000 per venue per year for high-volume operators, mostly pizza shops and delivery-heavy Asian or Indian concepts on Super Bowl weekends and Friday rushes. The U.S. restaurant industry as a whole leaves about $20 billion a year on the table to unanswered phones according to industry reports. Your number sits somewhere on a curve between $50,000 and $300,000, depending on phone volume, ticket size, and how many calls land in the 5pm to 8pm window.

Why does callback rescue not appear in the formula?

Because it almost never happens. Survey data shows 80 percent of callers do not leave voicemails and 85 percent will not attempt a callback. They open the next nearest restaurant in their delivery app and order there. So a missed call rounds to a lost order, with a small fudge factor for the 15 percent who try again. Adding a callback recovery term to the formula moves the final number by only a few percent, which is why most operators omit it for simplicity.