PieLine is a 102-second call, not a pitch deck.
Every write-up about PieLine is a five-bullet feature summary. You can do better in 102 seconds: press play on the demo call PieLine ships on its own landing page. This is that call, annotated second by second, with the exact moment PieLine takes a thinking beat, chains three modifier questions, attempts an upsell, and closes the ticket at $34.11.
The call that anchors this page
The hero on aiphoneordering.com plays a real recorded call. It is 102.36 seconds long, sampled at 60 Hz, with dual-channel amplitude envelopes (customer and AI on separate tracks so the waveform can show who is speaking). The transcript was produced by Deepgram’s multichannel model and checked into the repo. Everything that follows on this page is drawn directly from that transcript. No reconstruction, no paraphrase.
The anchor artifact
public/audio/dennys-order.mp3 plus src/components/voice-activity-data.ts
The JS file exports a typed VoiceDataobject with duration, sample rate, two amplitude envelopes (one per speaker), and 38 timestamped caption segments. The Denny’s order runs from turn 1 (“Hi. This is Denny on a recorded line.” at 0.00s) to turn 38 (“Thanks. Bye.” at 101.81s). Most other pages about PieLine do not mention this file exists, because writing about it requires opening the repo.
Beat 1, 0.00s to 9.36s: the opener and the order
The agent opens with a three-part greeting that ends in an open prompt, not a menu read-out. It says “Hi,” discloses that the line is recorded, then asks “What can we get for you?” Rob starts talking two seconds later and puts in two items in one breath: a Lumberjack Slam and a Coke. The agent does not interrupt. It lets the order land.
What the agent is doing here
- Open prompt, not a menu tree. “What can we get for you?” is a single request that invites a full order. No “press 1 for takeout.”
- Compliance one-liner. “on a recorded line” lands before the caller says anything material, which keeps the recording admissible and the restaurant covered.
- Barge-in tolerance. Rob begins “Hi. Yeah” before the greeting has fully settled. The agent waits it out, accepts the real request, and moves on.
Beat 2, 10.96s to 25.66s: the thinking beat, then three chained questions
This is the part most scripted demos do not show you, because it doesn’t look impressive until you realize what it is doing. At 10.96s the agent says “One moment, please.” Five seconds of quiet. Then at 15.98s it returns with a batch of three modifier questions, chained into one turn, for the Lumberjack Slam:
Three real things are happening in those 15 seconds. First, the agent looks up the Lumberjack Slam in the menu ontology and resolves which modifiers are required. Second, it batches the required questions into one turn instead of a back-and-forth ping-pong that would blow out the call length. Third, it explicitly reads back the default for the easy item (Coke) without asking Rob to re-specify it. That is how the call stays under two minutes end to end.
Beat 3, 29.39s to 47.57s: read-back, name capture, and a clean close
Rob answers the three modifier questions in the order they were asked. The agent read-backs the full line item (Lumberjack Slam, scrambled eggs, sourdough bread, soft drink), asks if there is anything else, and accepts a name (“Rob”) in place of a phone number. It could have closed the call right there at 47.57 seconds. It did not. What it did next is the single highest-ROI second of the conversation.
Read-back uses ontology fields, not the raw caller phrase
Rob said 'sourdough bread' and 'scrambled.' The agent replied with the canonical modifier names used by the POS item ID map, so the kitchen ticket will print the same words the cooks are used to seeing.
Name capture, no phone entry
'Put it under the name Rob' is the mobile-era equivalent of a handoff slip. The agent treats the name as order metadata and never asks for a phone number the caller already gave to the carrier.
Explicit 'anything else'
The agent gives Rob a second-to-last exit before the upsell, so the upsell does not feel like a stall. Consent to continue is what makes the next beat work.
Beat 4, 51.56s to 68.30s: the upsell, and why it actually lands
At 52.5 seconds the agent attempts the single interaction every restaurant owner cares about most. The exact wording is in the transcript, and the wording is the whole point:
Three things you cannot fake in a scripted demo. One, the upsell references the cart Rob already built (“might make your Coke jealous”) which you can only do if you still have the order state loaded. Two, Rob says yes, then immediately requests an add-on modifier (“strawberries, if that’s an option”). That is exactly the “cuisine-specific customization” the product page talks about in abstract, and here it is concrete: a strawberry topping modifier on a cheesecake SKU. Three, the phrasing is warm. “It’s so good.” That is not what a generic voice agent sounds like.
Beat 5, 71.34s to 99.28s: confirmation, placement, and the ticket
The agent repeats the full order back one more time, item by item, modifier by modifier, and asks “Is that correct?” Rob confirms. The agent then says out loud that it is placing the order (“Placing your order now”), reports completion (“Done”), gives the total (“$34.11”), and a pickup time (“12:45 AM”). Below is the message exchange the same way our engineering team would draw it on a whiteboard.
Rob's order, 102 seconds end to end
The full transcript, 38 turns
Every timestamp below is from the Deepgram transcription in the repo. If a moment on the recording contradicts a sentence on this page, the recording wins.
Live at these restaurants
PieLine is deployed and answering phones at real shops. These are the names on the landing page, not a hypothetical roster.
How PieLine gets from “sign up” to this call
The Denny’s call is beat-perfect because the work happened before the phone rang. Onboarding is hands-off for the owner and fits inside a day.
Connect the phone line
Forward your restaurant number to PieLine, or set PieLine as the overflow when staff cannot pick up. About ten minutes of phone-carrier config.
Menu scrape and POS mapping
PieLine's onboarding team crawls your existing online ordering page, maps every item to a POS item ID in Clover, Square, Toast, NCR Aloha, or Revel, and writes per-dish descriptions covering spice, sweetness, ingredients, and allergens.
Modifier tree wired
Half-and-half pizzas, protein substitutions, custom sushi rolls, eggs-and-bread pickers like the Lumberjack Slam, and strawberry-topping add-ons are declared as typed options on the POS item IDs. This is why the Rob call never improvises.
Go live, same day
PieLine starts answering calls. Active call monitoring and AI refinement run during the first 30 days. Escalated calls turn into the next week's ontology updates.
What changes when PieLine takes the phone
The numbers PieLine publishes are concrete and checkable against the product pages. They come from the Mylapore deployment and from the product’s own operating envelope.
Projected additional revenue per Mylapore location per day, from eliminating the missed-call bottleneck. Publicly endorsed by owner Jay Jayaraman.
Concurrent call slots on a single PieLine line. Friday dinner and game-day volume lands without hold music.
Share of calls handled end-to-end by AI at Idly Express. Edge cases route to a human with the full transcript attached.
PieLine vs. the obvious alternatives
Same problem, three different shapes. Here is how the Rob call would have gone on each.
| Feature | Alternative | PieLine |
|---|---|---|
| Picks up during Friday rush | Dedicated phone host ($3-4k/mo): one call at a time, calls in sick on busiest nights | 20 simultaneous calls, zero hold time, 24/7 |
| Chained modifier questions like the Lumberjack Slam | Per-minute voice service: handles the call but cannot look up required mods against a POS ontology | Required modifiers read off the dish ontology before the caller has to repeat themselves |
| Upsell with cart awareness | Script-driven IVR: reads a generic upsell line, no reference to what is already in the order | 'Might make your Coke jealous,' references the item already on the ticket |
| Ticket into POS without re-keying | Online ordering platform: only captures callers who already switched to the app | Clover, Square, Toast, NCR Aloha, Revel, 50+ POS integrations |
| Time to first answered call | New hire: 2-4 weeks of training, then re-training on turnover | Same-day go-live on most shops |
| Pricing predictability | Per-minute billing: bills spike on busy weeks | $350/mo up to 1,000 calls, $0.50/call after |
If your phone rings once an hour and never overlaps, a human host is fine. If your phone rings 12 times at 6:45 PM on a Friday, PieLine is the shape that matches the workload.
Hear a call like Rob's, on your menu, in 15 minutes
We scope your onboarding and set pricing on the same call. Live in under 24 hours on most POS systems.
Frequently asked questions about PieLine
What is PieLine?
PieLine is a 24/7 AI phone answering service built for restaurants. It picks up the phone, takes food orders in natural conversation, answers menu questions, handles reservations, upsells, and drops the completed ticket directly into the restaurant's point of sale. The same agent handles up to 20 simultaneous calls and is live in under 24 hours for most shops.
How does the PieLine demo call actually sound?
The demo on the PieLine landing page is a real 102.36-second Deepgram-transcribed call. A customer named Rob calls Denny's, orders a Lumberjack Slam and a Coke, gets asked three clarifying questions about eggs and bread, accepts a cheesecake upsell with strawberries, and hears the order confirmed back with a $34.11 total and a 12:45 AM pickup time. The audio and the transcript both live in the repo (public/audio/dennys-order.mp3 and src/components/voice-activity-data.ts).
What happens on a PieLine call that you cannot get from a script demo?
The agent takes a deliberate thinking beat. In the Rob call it says 'One moment, please' at 10.96 seconds, then comes back at 15.98 seconds with three modifier questions chained together: how would you like your eggs cooked, what kind of bread, white, brown, multigrain, or sourdough. You only see that cadence on a recorded call, not on a scripted marketing animation.
Does PieLine upsell, and what does that sound like?
Yes. At 52.5 seconds into the demo call, after the customer says 'put it under the name Rob,' the agent asks: 'Before I finish up, would you like to add a sweet treat like a slice of New York style cheesecake? It's so good. It might make your Coke jealous.' The customer accepts and asks for strawberries on top. That extra modifier lands in the ticket cleanly, which is why PieLine reports 95%+ order accuracy on orders that pass through the POS item ID map.
What POS systems does PieLine push tickets into?
Clover, Square, Toast, NCR Aloha, and Revel are live out of the box. 50+ POS integrations in total. If your restaurant runs a POS with an API, PieLine can wire into it during onboarding. Orders land on the kitchen ticket with the right item ID, modifier IDs, and price, so there is no re-keying.
How is PieLine priced?
$350 per month covers up to 1,000 calls. $0.50 per call beyond that. Onboarding is hands-off for the restaurant owner and included in the price. A 30-day money-back guarantee applies if the 95%+ order accuracy or the 90%+ AI-handled call rate does not hold at your location. Compared to hiring a dedicated phone host at $3,000 to $4,000 a month, that is a 70 to 80% cost reduction with 20x the simultaneous call capacity.
Who is already using PieLine?
Mylapore, an 11-location South Indian chain in the Bay Area, is rolling PieLine across all locations. Owner Jay Jayaraman publicly endorses it on and projects $500 per location per day in additional revenue. Idly Express (Almaden) is live and says 90%+ of calls are handled end-to-end by AI. Amber India and a few others are currently onboarding. PieLine is backed by Founders, Inc.
What happens when the agent cannot handle a call?
It transfers to your staff with a full transcript of the conversation so far. Complaints, catering requests above a threshold, novel menu items that are not in the ontology yet, and hard allergen questions on unverified dishes all route to a human with context attached. The agent is designed to refuse to guess on facts that could get the restaurant in trouble.
Where do I hear the demo for myself?
Go to aiphoneordering.com and press play on the voice activity clip in the hero. The waveform you see moving is the exact dual-channel amplitude envelope from Deepgram's multichannel transcription, sampled at 60 Hz. The captions below it are the same transcript this page walks through.
How did this page land for you?
React to reveal totals
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.