The short answer
Yes — an AI receptionist can quote real prices from your own price list on a live call, and the good ones do it in the first ninety seconds. The AI looks up the caller's job in a pricing catalog you control, applies your service and travel fees, and speaks one all-in number. As of July 2026, this is the single clearest dividing line between an AI phone agent that books work and one that just takes messages: the first quotes, the second promises a callback.
That distinction matters more than it sounds. A caller who hears "that'll be $295 all in, and I can get a tech out this afternoon" is a caller you can close on the call. A caller who hears "let me have someone get back to you with pricing" is a lead you now have to chase, and they are already dialing the next result. Everything below is how the quoting actually works under the hood, what it can and cannot price, and the guardrails that keep it from costing you money.
Why "we'll call you back with a price" loses the job
Price is usually the second thing a service caller says, right after the problem. They have a broken thing and a budget question, and both arrived at the same moment. If your phone answers that question, you're in the running. If it doesn't, you're a message on a list.
The old model handled this with a human who knew the prices. That works until the human is on another call, on a job site, asleep, or new. Then the price question either goes to voicemail or gets answered wrong. Traditional answering services are worse: the operator has no price list and no authority, so every call ends in a message. That's the structural gap covered in answering service vs. AI receptionist.
The cost is easy to see once you count it. Take a business fielding 200 calls a month where 120 are real service inquiries. If quoting on the call converts even fifteen percentage points better than a callback promise — a modest assumption, not a study — that's 18 extra booked jobs a month. At a $400 average ticket, that's $7,200 in monthly revenue that hinged on whether the phone could say a number. Run your own version of that arithmetic against your ticket size; the conclusion rarely changes.
How catalog-based quoting actually works
There are two ways an AI can produce a price, and only one of them is safe.
The unsafe way is letting the language model estimate. Ask a general-purpose AI what a service costs and it will produce something plausible-sounding drawn from the open internet. That number is not yours. It will drift between calls, it will not match your market, and it will occasionally be wildly wrong in the direction that costs you a job or costs you margin.
The safe way is a lookup. Your prices live in a structured catalog — a table of services, job types, and the variables that change the price. When a caller describes their job, the AI extracts the relevant details, queries the catalog for a match, and speaks back what the catalog returned. The model handles the conversation; it does not handle the pricing. That separation is the whole design.
The lookup is layered, because real callers give partial information:
- Exact match. The caller's job maps to a specific catalog row. This is the best case and the most common one once your catalog has depth.
- Broader match. The caller gave less detail than the catalog needs, so the system matches on a wider category and quotes the representative rate for it.
- No match. Nothing in the catalog covers it. The AI must not invent a number here. It captures the details, says a technician will confirm the exact price, and books or queues the follow-up.
That third branch is the one that separates a system you can trust from one you can't. Silence on an unpriced job is honest. A guess is a liability.
Fees, travel, and the mid-call price jump
Here's the failure mode that quietly kills more bookings than a high price ever does: the AI quotes $295, the caller agrees, and then thirty seconds later the AI adds a $100 after-hours fee. The caller doesn't hear a fee. They hear a bait and switch, and they hang up.
The fix is architectural, not conversational. Any fee that applies — a mobile or trip charge, mileage-based travel cost, an after-hours or weekend surcharge, a card processing fee — should be computed by server-side rules and folded into a single total before the AI ever says a number. The caller hears one figure, once: "$390 all in, no other fees." No arithmetic happens in the model's head, because language models are unreliable at arithmetic under conversational pressure and they will occasionally quote the pre-fee number.
Three rules that hold up in production:
- One number, all-in. Compute service price plus every applicable fee, then speak the total. Never speak a base price and add to it.
- Time-of-day rules run on your clock, not the server's. An after-hours surcharge evaluated in the wrong timezone will bill a two o'clock afternoon call as after-hours. This is a genuinely common bug and it makes the business look dishonest.
- Fail open on fees. If the fee engine can't resolve — bad config, missing timezone data — quote without the surcharge rather than guessing at one. Undercharging by $100 on rare calls is cheaper than reputational damage on every one.
Quote integrity: the same customer, the same price
A caller who was quoted $295 last Tuesday and hears $420 today will not conclude that your prices went up. They will conclude you make prices up. This is the most damaging quoting bug there is, and it's entirely preventable.
The mechanism is caller history. Before quoting, the system looks up the inbound number and pulls what it already knows: previous quotes, past jobs, open appointments, the vehicle or property on file. If a prior quote exists for the same job on the same asset, the AI honors it. If a correction is genuinely unavoidable — the caller described the job differently the first time — it says so explicitly rather than quietly re-quoting.
That same history lookup does more than protect prices. It's what lets the AI open with "I've got you at 4400 Oak — is this about the same unit?" instead of interrogating a returning customer from scratch. It's also what makes it possible to confirm an existing appointment when someone calls asking where the tech is, instead of the far worse "I don't see anything under this number." The dispatch side of that record is covered in CRM and dispatch software for multi-tech businesses.
What a quoting AI can and can't price
Being honest about the boundary is what makes the whole thing usable.
| Job characteristic | AI quotes it directly | Why |
|---|---|---|
| Standard service, known category | Yes — exact catalog price | The catalog has a row; the lookup returns it |
| Priced by a variable the caller knows (rooms, feet, units) | Yes — computed from your rate | Caller supplies the variable, rules do the math |
| Standard job outside business hours | Yes — all-in with surcharge | Fee engine adds it before the number is spoken |
| Job needing a visual inspection | Range only, then site visit | Nobody can price unseen damage over the phone |
| Custom or bid work | No — capture and hand off | The number depends on judgment, not a table |
| Unusual model, rare equipment, edge case | No — technician confirms | Guessing here costs margin or costs the job |
The pattern: anything your own dispatcher could quote from a laminated sheet, the AI can quote. Anything that requires a person to look at it, the AI should capture and route. A system that pretends otherwise is worse than one that admits the limit.
Booking is the point, not the quote
A quote is only valuable if it turns into a job on a calendar. The strongest version of this flow never separates the two: the AI quotes, gets agreement, and books the appointment in the same conversation — same call, no callback, no "someone will reach out."
That requires the AI to actually hold the calendar, not just describe it. It needs the real availability, the right technician, and the ability to write the appointment. When those live in the same system, the caller hangs up with a confirmed time and an SMS confirmation already sent. When they don't, you get the familiar gap where a quote was given, nobody wrote it down, and the job evaporates. The mechanics are in how AI appointment booking works.
Two details that matter operationally:
- Ask the time before booking. If the AI books "ASAP" before the caller says "actually, one o'clock works better," you've created a wrong appointment that someone has to fix. The requested slot has to be captured first.
- Confirm the total in the confirmation. The SMS should restate the all-in price and the window. That single message eliminates most day-of disputes and most no-shows. More on that in reducing no-shows with appointment reminders.
Quoting in Spanish, at 2 a.m., during a surge
The reason quoting has to be automated rather than staffed is that price questions do not respect business hours or language. A meaningful share of service calls in most US markets arrive in Spanish, and a meaningful share arrive when nobody is at a desk.
An AI receptionist quotes identically at 2 a.m. and 2 p.m., in English or Spanish, whether it's one call or twenty simultaneous ones. That last property is the underrated one. A storm, a heat wave, or a well-timed ad flight produces a call spike that no staffed phone can absorb, and the calls you drop during a surge are your most valuable ones because demand and intent are both peaking. The AI answers all of them concurrently and quotes all of them the same. That's the case made in handling seasonal call volume and, for the language side, bilingual Spanish answering. The after-hours share alone is usually larger than owners expect — see the after-hours calls playbook.
Keeping the catalog honest
The catalog is the product. Everything above depends on it being right, and a stale catalog is the one way this fails badly at scale.
Load real prices, not seed data. A catalog populated with placeholder rows will happily quote placeholder prices with total confidence. Every row should be one you'd stand behind.
Watch for the fallback tell. If the same number shows up on every quote regardless of job, your lookup isn't matching and you're quoting a default. That's a config bug wearing a price tag, and it's invisible unless you sample real calls.
Update in one place. A rate change should take effect on the next call, everywhere — the phone, the booking confirmation, the invoice. Prices maintained in three systems drift within a month.
Sample your calls weekly. Pull a handful of transcripts and check what was actually quoted against what should have been. Call recording, transcription, and lead scoring on the Pro plan make this a ten-minute review instead of a project. The attribution side of that data is in call tracking and attribution.
Where quoting sits in the wider stack
Quoting on the phone is one function, and it only pays off when the surrounding steps exist. The catalog feeds the quote; the quote feeds the booking; the booking feeds dispatch; dispatch feeds the invoice; the invoice feeds the review request. Break any link and the value leaks out at that joint.
That's the argument for running these as one system rather than five subscriptions stitched together with exports. When the phone, the calendar, the CRM, and the POS share one record, the price a caller heard is the price on their invoice without anyone retyping it. When they don't, someone re-enters it and eventually gets it wrong. The full comparison is in all-in-one vs. point solutions, and the payment end is in getting paid faster with payment links.
For the record layer that makes all of it queryable in plain language — "what did we quote that number last month?" — see Jarvis AI Brain.
A four-week rollout that doesn't blow up
Businesses that get this working share the same sequence, and it isn't "turn it all on Monday."
Week one — catalog only. Load your real prices for the jobs that make up the bulk of your volume. Not everything you've ever done; the twenty or thirty rows that cover most calls. Leave everything else unpriced deliberately, so the AI hands off rather than guesses.
Week two — quote, don't book. Let the AI quote and capture, but route the booking to a person. You get the price behavior under observation without the calendar risk. Pull ten transcripts at the end of the week and compare what was quoted against what should have been.
Week three — fees and edge cases. Configure travel, after-hours, and any surcharge rules. Test each one at the boundary: a call at 6:59 p.m. and one at 7:01 p.m. should differ by exactly the surcharge and nothing else. Timezone bugs surface here or they surface on a customer.
Week four — hand it the calendar. Turn on booking, keep sampling calls weekly. By now you know what the phone says, so the only new variable is whether the slot it picks is the slot the caller asked for.
The failure pattern is the inverse: everything on at once, nobody reads a transcript for a month, and a lookup bug quotes the same default number on 400 calls before anyone notices.
What it costs
Run with Jarvis includes catalog-based quoting on every plan, because quoting is not a premium feature — it's the reason the phone is worth automating.
- Core — $500/mo. 500 AI call minutes, $0.45/min after. AI receptionist 24/7 in English and Spanish with quoting and booking, calendar, GPS tracking and route optimization, ETA and arrival SMS, POS, invoicing, QuickBooks sync, CRM, review automation, chargeback defense.
- Pro — $750/mo. 1,000 minutes, $0.40/min after. Everything in Core plus call tracking with DNI, Google Ads and Meta attribution, transcription, lead scoring, sentiment and intent analysis, call recording and playback — the layer that lets you audit what was actually quoted.
- Elite — $1,200/mo. 2,500 minutes, $0.35/min after. Everything in Pro plus AI campaign building, Google Business Profile management, AI review replies, competitor intelligence, and the Jarvis AI Assistant.
Zero setup fees, unlimited users, month-to-month on all three. Full detail at pricing, the wider cost picture in what AI operations actually cost, and the return math in AI receptionist ROI.
The bottom line
An AI receptionist can absolutely quote from your price list, and if it can't, it isn't finished. The requirements are specific: a real catalog you control, a lookup rather than a guess, server-side fee math producing one all-in number, caller history so repeat customers hear a consistent price, an honest fallback on anything unpriced, and a booking step attached to the quote so the answer turns into a job.
Get those six things right and the phone stops being a message-taking machine. It becomes the thing that answers the question every caller actually leads with — and then puts the work on the calendar before they hang up. Ready to load your catalog? Get in touch.
