POS / channel design

Self service POS: the kiosk is one form factor, the phone is the other

Every POS vendor markets self service as a touchscreen kiosk. That is one form factor. There is a second one most operators already have wired up and ignore: the phone, answered by an AI agent that POSTs the order directly to the same POS endpoints the kiosk uses.

Matthew Diakonov, Written with AI

Published May 10, 20269 min read

Direct answer (verified 2026-05-10)

A self service POS is any system that lets a customer place and pay for an order without staff between them and the point of sale. Vendors like Square and Toast define the category as a touchscreen kiosk; the same definition applies to an AI phone agent that POSTs orders directly to the POS using the same item and modifier identifiers a kiosk would write. The form factor is different, the order path is structurally the same.

Definition cross-checked against Square's self-service kiosk explainer and Toast's restaurant kiosk page.

What "self service" actually means at the POS layer

Walk into a Toast or Square sales conversation about self service and you will be shown a kiosk. The pitch is consistent: a 21-inch touchscreen the diner pokes through, a card reader stuck to the side, a small printer underneath, and a stand that bolts to the counter. The marketing line is that customers self-order, staff stop taking orders, average ticket goes up by 12 to 30 percent because the diner is not rushed by a queue behind them.

All of that is true. None of it is the definition of self service. The definition is operational, not visual: the customer is the only human in the loop on the order, and the POS receives a fully formed ticket without anyone retyping it. The kiosk satisfies that definition through a touchscreen. Anything else that satisfies it through a different interface is also self service POS.

That second qualifier (no staff retype) is what separates a self service POS from an order-management overlay. A DoorDash tablet that beeps and waits for a host to type the order into the Toast terminal is not self service. The customer placed the order on a phone, but a staff member is still the bridge to the POS. Same goes for an AI voice product that emails the staff a transcript and trusts them to enter it manually. The fact that the customer talked to a machine does not promote the order to self service if a person on the restaurant side has to retype it before the kitchen sees it.

The order path is structurally the same on both channels

Strip the form factor away and look at what the kiosk does after the diner taps Confirm. It assembles a cart in the local kiosk app, then POSTs to the POS's native order-creation endpoint with item GUIDs from the POS menu sync, modifier GUIDs, totals, taxes, and a customer tag. The POS responds with an order id, fires a ticket to the kitchen display, and the kiosk reads the order number back to the diner.

Kiosk flow: diner taps to confirm

Now look at the same path on the phone channel, with PieLine in the kiosk-app slot. The agent assembles a cart from the same POS menu sync, maps the caller's utterances to the same item and modifier GUIDs, and POSTs to the same order-creation endpoint with the same payload shape. The POS responds the same way. The kitchen display cannot tell the two tickets apart.

Phone flow: caller finishes ordering

The interface differs, the payload does not. The defining property we set out at the top (customer in direct contact with the POS, no staff retype) is satisfied on both. That is what makes the phone channel a self service POS channel when the agent commits properly, and what makes it a glorified answering service when it does not.

What an operator actually thinks vs. what the POS sees

Most restaurant operators have an intuitive map of which channels are self service and which are staff-mediated. The map is wrong on the phone column.

The mental model vs. the POS reality

Self service is the kiosk in the dining room. Phone orders are inherently a staff job because someone has to pick up. The POS doesn't really care which channel an order came from as long as the cashier types it in correctly.

Self service = kiosk only
Phone is always a staff cost
Channel does not show up in POS data

What restaurants pay per channel for self service

The cost shape of a kiosk and the cost shape of an AI phone agent are both subscription-on-top-of-something, but the something is different. A kiosk is hardware first, software second. PieLine has no on-premise hardware; the cost is a flat monthly that scales with call volume. Comparing the two side by side, with prices verified against vendor pages this month:

Feature	In-store kiosk	AI phone agent (PieLine)
Customer is in direct contact with the POS	Yes, via touchscreen	Yes, via voice
Order arrives at kitchen with no staff retype	Yes	Yes
Item GUIDs and modifier GUIDs written from POS catalog	Yes (kiosk app reads from POS menu sync)	Yes (agent reads same menu sync, maps utterances to GUIDs)
Hardware per unit	$149 (Square) to $999+ (Toast Flex) plus stand, printer, card reader	None on premise; phone line forwards to PieLine
Recurring software cost	$30 to $69 per kiosk per month, per device	$350 per month for 1,000 calls, $0.50 per call beyond
Concurrency	One customer per kiosk; operators run 2 to 4 to handle a rush	Up to 20 simultaneous calls on a single forwarded line
Hours of operation	Open hours only (locked when restaurant is closed)	24/7, including after-hours pickup orders for next-day fulfillment
Card present versus card not present	Card present (chip, tap), 2.49% + $0.15 typical	Card not present, 2.9% + $0.30 typical, higher chargeback risk
Channel share at most takeout-heavy restaurants	Captures dine-in walk-ups: 10 to 30% of orders	Captures phone takeout and delivery: 20 to 40% of orders

The two columns are not substitutes. They are complements. A kiosk captures dine-in walk-ups; an AI phone agent captures phone takeout and delivery. Most restaurants we talk to have one channel covered and one wide open. The pattern is usually that bigger chains have kiosks and no phone strategy beyond rollover-to-voicemail. Smaller independents staff the phone manually and never considered a kiosk. Whichever channel is open is the one leaking calls or orders.

What this looks like in production at one chain

Mylapore is an 11-location South Indian chain in the Bay Area, which we have written about elsewhere. They started with the phone channel (not the dining room) because phone takeout was the largest leaking channel at peak. After the AI phone agent went live, the orders arrived in their POS as native tickets, indistinguishable from dine-in tickets at the kitchen, with the source field set to phone. Two cashiers at the San Jose location were redeployed to expediting and prep.

“The experience was better than speaking to a human. No hold time, no confusion, no rushing.”

Customer feedback

Mylapore South Indian, Bay Area

That quote is from a diner, not the operator, and it captures the thing that makes this category work: the customer does not feel they traded down to a self service experience. From the caller's perspective, every call is answered, every order is read back, and the line is never busy on a Friday at 7pm. From the POS's perspective, a phone ticket is now identical in shape to a kiosk ticket, which means revenue per channel and labor cost per channel are finally measurable.

Where the form factors honestly diverge

None of the above means the kiosk and the AI phone agent are interchangeable. They are equivalent at the POS-commit layer and different at the channel layer in ways that matter.

Card present vs. card not present. The kiosk takes a chip card or NFC tap on premise; the phone channel takes a card not present and pays a higher processing rate. That is a fixed cost per transaction the operator should budget into the channel. For pickup and delivery, this is not new. Phone takeout has been card not present for thirty years.
Visible modifier discovery. A kiosk shows a diner the modifier list as buttons. The phone agent has to know to ask. Half-and-half pizzas, spice levels, protein substitutions, and custom rolls only work if the menu was configured with those modifier paths during onboarding. This is the work most generic voice products skip and where cuisine specialization actually matters.
Error recovery. A diner who hits a dead end on the kiosk can typically hit Back and try again. A caller who hits a dead end needs a graceful transfer to a manager with full call context. The presence or absence of that transfer is the thing to test on any AI phone vendor before signing.
Accessibility cuts both ways. Voice is friendlier for older customers and customers with low vision; touch is friendlier for customers with hearing loss or non-native speakers reading rather than listening. The once-and-for-all answer is to run both. Self service is not a binary you flip on for the restaurant; it is a per-channel capability.

See what a phone-channel self service POS looks like on your menu

Fifteen minutes. We walk through the order-commit path on your POS, including modifier GUID mapping for the items most likely to break.

Frequently asked questions

What is a self service POS, in one sentence?

A self service POS is any system that lets a customer place and pay for an order without staff between them and the point of sale. The recognizable form factor is a touchscreen kiosk in the dining room. A second form factor, less recognized, is an AI phone agent that POSTs the order directly to the same POS endpoints a kiosk would use. Both share the defining property: the customer is in direct contact with the POS, and no staff member retypes the order before it reaches the kitchen.

Why do POS vendors only show kiosks when they market self service?

Two reasons. First, kiosks are hardware vendors sell, with a per-unit margin and a recurring per-device software fee. The Square Kiosk is $149 of hardware plus $50 a month per active device. The Toast self-ordering kiosk is $999 plus a $69 monthly software fee. There is no equivalent SKU for the phone channel because the phone is not POS hardware; the POS just exposes an order-creation API. Second, the kiosk is visible. An AI phone agent that commits to the POS looks like an extra phone line to the operator. The order arrives in the kitchen identically to one a cashier typed in. The only operator-facing artifact is a row in the orders table.

Is taking a phone order really self service if the customer is talking to a voice agent?

Self service is defined by who is in the loop on the order, not by whether the interface is silent. A diner tapping through a kiosk is having a structured conversation with the POS via a touchscreen. A caller speaking to a voice agent that maps utterances to menu item GUIDs is having the same structured conversation via audio. Neither has a staff member typing on their behalf. The criterion that matters at the POS layer is whether anyone retypes the order before the kitchen sees it. If the answer is no on both channels, both are self service.

How does payment work over the phone if the kiosk takes a card swipe?

The card-present versus card-not-present distinction is what splits them, and it has nothing to do with self service status. A kiosk reads a chip card or NFC tap on premise, which clears as card present. An AI phone agent collects the card over the phone and runs it as card-not-present through the POS or a payment processor the POS exposes. The processing rate is higher (usually 2.9 percent plus a per-transaction fee instead of 2.49 percent) and chargeback risk is higher, but for delivery and pickup orders it is the same way takeout has been processed for thirty years. The order itself still flows into the POS as a normal ticket.

What share of restaurant orders is the phone channel actually worth bringing into self service?

It depends on cuisine and format. For pizza shops, Indian restaurants, Chinese restaurants, Mexican restaurants, and family restaurants that take takeout and delivery, phone is commonly 30 to 40 percent of orders, sometimes higher in markets with older demographics or weak app penetration. For a fast casual that has pushed customers to its mobile app, phone might be 5 percent. Operators should pull a week of POS data and look at the source field on each ticket: phone, web, dine-in, drive-thru, third-party. Whatever percentage shows up next to phone is the channel currently being staff-mediated, and the channel an AI phone agent would convert to self service.

What happens to staff if both the dining room and the phone go self service?

The pattern most operators are running, including the Mylapore chain (eleven South Indian restaurants in the Bay Area) we work with, is to redeploy. Mylapore eliminated the need for two cashiers at its San Jose location after switching the phone channel to AI and moved those people into expediting, prep, and supporting the rollout to new locations. The labor saved on order taking does not disappear from the restaurant, it shifts to roles that actually need a person. The math that matters is labor-per-cover and orders-per-shift, not headcount in a vacuum.

Where do an AI phone agent and a kiosk genuinely diverge?

Three places. First, error recovery: a kiosk diner who hits a dead end can usually figure out the touchscreen, but a caller who gets confused needs the agent to gracefully transfer to a human with full context. Second, accessibility: voice is friendlier for older customers and customers with low vision; touch is friendlier for customers with hearing loss or non-native speakers reading rather than listening. Third, modifications: a kiosk presents the modifier list as visible buttons, an AI phone agent has to know to ask 'what spice level' or 'half and half' without prompting. The first two are properties of the channel that an operator should embrace by running both. The third is a property of how the agent's menu was built; cuisine-specific configuration matters here and is the work most generic voice bots get wrong.

Adjacent guides