Operations of a restaurant: inside every phone order is a 14 second window where average ticket is decided, and we have the timestamps to prove it

Average ticket (AOV) is one of the four numbers every general manager looks at every week, alongside food cost percent, labor cost percent, and customer count. Service has had a lever for it for a hundred years (the server at the table, the host upsell line, the dessert menu drop). Online ordering inherited the lever in the 2010s (the suggested-add modal at checkout, the cart upsell). The phone never had a comparable lever, because phone upsells depended on a tired cashier remembering to ask, in a noisy kitchen, on a busy Friday, and producing zero artifact when they did or did not. AOV on the phone row was a vibes number.

On PieLine's reference order call, the customer closes the base order at 49.165 seconds with the line 'No, that's it, you can put it under the name Rob.' The AI then has a discrete operational slot before the order is committed to the POS. That slot starts at 52.525 seconds, when the AI says 'Before I finish up, would you like to add a sweet treat like a slice of New York style cheesecake?' and ends at roughly 67 seconds, when the customer has accepted and added a strawberry modifier. The base order is locked, the confirmation has not yet started, and AOV is being decided. That is the 14.48 second window, the same way a kitchen ticket has a fire-to-handoff window.

No. The reference call is one real recording (committed at /public/audio/dennys-order.mp3, transcribed at /src/components/voice-activity-data.ts), not a synthetic average. Different calls produce different windows depending on order complexity, modifier density, and how decisive the caller is. The operationally interesting fact is not that 14.48 seconds is universal, it is that the window now has a beginning, an end, and a recorded outcome on every single call. Operations can build a distribution from those windows the same way they build a kitchen-ticket-time distribution.

Because for most of the last twenty years, the phone could not produce a record that supported it. Operations guides only list metrics that have a backing artifact. POS reports give AOV by check, by daypart, by station. Online platforms give AOV by checkout. The phone produced a carrier line item with no per-call breakdown, so AOV was either pooled with walk-up tickets at the register or invisible. When PieLine writes the order into Clover, Square, Toast, NCR Aloha, Revel, or any of about 45 other systems, the phone-channel AOV finally lands on the same report as the dine-in AOV, with a transcript attached for audit.

Yes. Open /src/components/voice-activity-data.ts in the aiphoneordering.com repository. The captions array is the structured transcript. Find the entry where speaker is 'ai' and start is 52.525, the text is 'Before I finish up, would you like to add a sweet treat like a slice of New York style cheesecake?' Find the entry where speaker is 'customer' and start is 62.465, the text is 'Sure. Yeah. I'll get one, slice of, the cheesecake.' These are not summary numbers; they are subsecond timestamps written by Deepgram's multichannel transcription and committed to the repo. The audio is at /public/audio/dennys-order.mp3, the generator is at /scripts/build-voice-activity-data.py.

Three concrete changes. First, the GM gets phone-channel AOV broken out as its own row, not pooled into the register. Second, the GM gets an upsell-attempt rate (was the prompt fired?) and an upsell-conversion rate (did the caller accept?) per location, the same shape they already use for dessert attach in the dining room. Third, the GM can audit any specific upsell by reading the transcript and the modifier capture, the way they would audit a comp by reading the slip. None of this required new dashboard software; it required a process artifact for the phone channel, which now exists.

It looks like a call where the AI fired the prompt and the caller declined. The window still appears in the record, with start and end timestamps, the prompt text, and the customer response. That is the value of having an artifact instead of an outcome: the prompt that did not convert is still measurable, which means operations can A/B the prompt language, the timing in the call, and the menu item being suggested, and watch the conversion rate move. Service has had this loop in the dining room for decades. The phone is just catching up.

It does not replace any of them; it gives the operations dashboard a row that used to be blank. Purchasing has a buyer and an invoice. Inventory has a counter and a count sheet. Prep has a line cook and a kitchen ticket. Labor has a scheduler and a punch. Service has a server and a check. The phone channel now has an AI agent and a turn-level transcript. The AOV lever is one specific cell in the new row, the one that ties the phone channel into the same revenue-per-customer conversation the rest of the floor already has every Monday.

Pricing is 350 dollars per month for up to 1000 calls and 50 cents per call after. A single-location pizza shop doing 200 phone orders a week (about 800 a month) lands inside the base tier. The threshold for caring about phone-channel AOV as an operational metric is roughly when phone orders are at least a few percent of total tickets; below that, the lever exists but the dollar impact is in the noise. Above that, every percentage point of upsell conversion is real money on the same scorecard line as your dine-in dessert attach.

PieLine reads the customer's confirmed order before generating the upsell prompt. On the reference call the caller had ordered breakfast (lumberjack slam) and a soft drink, so the AI offered cheesecake, framing it against the drink ('it might make your Coke jealous'). For a different base order the prompt is different. The operations consequence is that the upsell is order-aware by default, which is the same thing a server does in their head and the same thing a checkout-modal cannot do without a recommendation engine bolted on. The artifact lets you confirm this happened on any specific call by reading the transcript.

Same day for most setups. Forward the existing main line to PieLine (or set it as overflow), have the menu scraped from the existing online menu, map the POS (Clover, Square, Toast, NCR Aloha, Revel, or 45 others), and the first calls land as structured records within hours. The first weekly operations digest, with the upsell rate row included, lands the following Monday. The 14.48 second window only appears once you have real calls, and it appears on every one of them.