The short answer
As of August 2026, a full AI phone answering platform for a service business costs roughly $500 to $1,200 a month, and the honest range is that tight because the market has largely converged on flat monthly plans with an included pool of call minutes. Run with Jarvis sits at $500 for Core with 500 included AI call minutes, $750 for Pro with 1,000, and $1,200 for Elite with 2,500 — month-to-month, zero setup fees, unlimited users, no per-call and no per-booking charge, with flat published overage at $0.45, $0.40 and $0.35 a minute respectively.
That is the number. This article is about the thing that actually decides your bill, which is not the number.
Two vendors can both say five hundred a month and bill you three hundred dollars apart in a busy month, because they count different things. The pricing structure — per-minute, per-call, flat monthly with a pool, or per-seat — determines whether a good month costs you more or less, whether you can forecast at all, and whether the vendor's incentives point the same direction as yours. Below are the four structures, what each does to a real service business, and the five questions that strip a quote down to what it will actually cost you.
Why this is confusing on purpose
Pricing opacity is a sales tactic, and in this category it takes a specific form: a low headline number attached to a unit that is not the unit you consume.
Ninety-nine dollars a month is a real number that appears in this market. It usually buys a small block of minutes or calls, after which you are on a metered rate, and the metered rate is where the actual revenue lives. The advertised price is a doorway, not a bill. That is not fraud, and the terms are usually disclosed somewhere, but it does mean that comparing headline prices across vendors tells you almost nothing.
The fix is boring and effective. Convert every quote into the same currency: your expected total bill in a normal month, and your expected total bill in your busiest month. Two numbers per vendor. Everything below is in service of computing those two numbers.
The four pricing models
| Model | How it bills | Best for | The trap |
|---|---|---|---|
| Per-minute, pure metered | Every minute of every call | Very low or wildly erratic volume | No ceiling, and a good month costs more than a bad one |
| Per-call | A flat fee each answered call | Businesses with short uniform calls | You pay the same for a 40-second robocall as a 12-minute booking |
| Flat monthly with a minute pool | Fixed base, included minutes, published overage | Most service businesses | Only fair if the overage rate is published and flat |
| Per-seat or per-user | Per person with access | Office-heavy teams, rarely trades | Punishes you for giving technicians access to your own system |
Pure per-minute metered. You pay for what you use with no base fee. This is genuinely the cheapest option if you take fifteen calls a month, and it is the worst option the moment you grow, because your bill scales linearly with success and has no ceiling. There is a subtler problem: it makes you reluctant to let the system handle long calls, and long calls are the ones that book. Any pricing model that makes you hope for shorter conversations is working against your revenue.
Per-call. A flat fee per answered call, commonly a dollar or two. It sounds fair and is structurally poor for service work, because your calls are not uniform. A twelve-minute intake that ends in a booked $900 job and a forty-second wrong number cost you the same. In most service businesses a meaningful share of inbound calls are robocalls, wrong numbers and existing customers asking a one-sentence question, and per-call pricing charges you full freight for all of them. The vendor earns most on the calls worth least to you, which is an incentive mismatch you can feel over time.
Flat monthly with an included pool. A fixed base price includes a set number of minutes, and beyond the pool you pay a published flat rate. This is the model most service businesses should want, for one reason above all others: it is forecastable. You know your base, you know your pool, and you can calculate any month's ceiling before it happens. It also aligns incentives — the vendor is not paid more when your calls run long, so nothing in the system is nudging the conversation toward brevity at the expense of a booking.
The fairness of this model rests entirely on the overage rate. A published, flat rate that gets cheaper as you move up tiers is a good-faith structure. An unpublished rate, or one that escalates with usage, converts the plan into a metered plan with extra steps.
Per-seat. Common in general business software and mostly wrong here. Charging by user discourages you from giving your dispatcher, your office manager and your technicians access to the system that runs your jobs, which is precisely backwards. If a vendor prices this way, price the seats you will eventually need, not the ones you would grudgingly start with.
Running your own numbers
Estimating is a ten-minute job and it settles most of the decision.
Step one, get your average call length. Pull your last ten real service calls and time them. Do not use a vendor's assumption. A full intake — greeting, qualifying the problem, capturing the address and equipment, quoting from a price list, offering a slot, confirming — commonly runs four to six minutes. Emergency calls run shorter because the customer wants a truck, not a conversation. Quote-shopping calls run longer, which is its own topic and one we cover in handling price shoppers on the phone.
Step two, count answered calls. Your carrier's log has this. Count last month's inbound calls, and separate business hours from after hours, because the after-hours block is the volume a human answering setup was never touching and is therefore new minutes.
Step three, multiply. Calls times average length equals monthly minutes. Then map:
- At a five-minute average, 500 minutes covers about 100 calls a month, roughly three a day including weekends and the middle of the night.
- 1,000 minutes covers about 200 calls, six or seven a day.
- 2,500 minutes covers about 500 calls, sixteen a day.
At a brisker four-minute average those pools stretch to roughly 125, 250 and 625 calls. And in practice your real capacity runs higher than the arithmetic suggests, because robocalls, wrong numbers and one-question calls consume a fraction of a full intake.
Step four, price your worst month. This is the step people skip and it is the one that matters. Take your busiest month in the last year, add a margin, and compute base plus overage. On Core, 300 minutes over the pool is 300 times $0.45, or $135 on top of $500 — $635 all in. That is your ceiling, and you knew it before it happened. Being able to do that calculation at all is the argument for this pricing model.
Step five, read the overage as a signal. Consistent overage is not a penalty, it is your business telling you it outgrew the tier. On Core, roughly 550 overage minutes costs about $247, which is essentially the $250 gap to Pro — and Pro doubles the pool to 1,000 minutes and adds the entire call-tracking layer. When staying costs the same as upgrading, upgrade. The plan-choosing guide works that math through in detail.
The line items that are not the price
Headline price is one line. These are the others, and they are where quotes diverge.
Setup and onboarding fees. Common in this market, sometimes several hundred to a few thousand dollars. Run with Jarvis charges none on any plan. Ask directly and get it in writing.
Per-seat charges. Ask whether adding your dispatcher or a technician costs money. On the Run with Jarvis plans users are unlimited.
Per-booking or per-lead commissions. Some vendors take a cut per appointment booked. Consider what that means at scale: the more successful the system is, the more it costs, forever. There is no per-call or per-booking fee on any Run with Jarvis plan.
Integration fees. Charges to connect your calendar, your CRM or your accounting software. This tends to be the sharpest difference between a platform and a point tool, because a platform where booking, CRM, dispatch and invoicing are the same system has nothing to integrate. Our comparison of all-in-one platforms against point solutions covers that structural difference.
Contract length and early termination. Annual commitments are common. Every Run with Jarvis plan is month-to-month, which changes the risk profile of the decision — being wrong costs you one month rather than a year.
Number and telephony charges. Ask whether phone numbers, call recording and transcription are included or billed separately.
What you stop paying. The genuinely fair comparison subtracts the tools the platform absorbs. If you currently pay separately for scheduling software, invoicing software and a review-request tool, and the entry plan includes all three alongside a CRM, POS, GPS routing, ETA texts and QuickBooks sync, then the net cost is far below the sticker. Owners routinely forget to run this subtraction, and it is often several hundred dollars a month.
Comparing against a human answering service
Because this is the comparison most owners are actually making, one clarifying point on structure.
Traditional answering services typically bill per minute or per call, and the deeper issue is not the rate — it is what the money buys. A message service takes a message. The quoting, the booking, the CRM entry and the dispatch still land on your team the next morning, which means you are paying for capture and doing conversion yourself.
An AI platform at $500 a month answers around the clock in English and Spanish, quotes from your actual price list, books directly into your calendar, writes the customer into the CRM, dispatches the job, sends the ETA text, invoices it and requests the review. The comparison is not per-minute rate against per-minute rate; it is capability per dollar. We ran that comparison properly in AI versus human answering services and looked at the traditional-service side specifically in answering service versus AI receptionist.
There is also a coverage dimension no per-minute rate captures. A human service answers one call at a time; when three people call during a storm, two hear ringing. Concurrency has no line item on either quote and it is worth more than the price difference.
What actually moves your minute usage
If your bill is driven by minutes, it is worth knowing what drives minutes, because several of the levers are under your control and owners rarely realize it.
Your price list. This is the biggest one and it surprises people. When the system can quote a job directly, the call ends when the customer books. When it cannot — because the service is not priced, the vehicle or equipment is not in the catalog, or the rules for your service area are vague — the conversation turns into a negotiation about what happens next, and negotiations are long. A well-loaded price list is a minute-reduction project as much as it is a revenue project.
Your service-area rules. A caller forty miles outside your radius should be identified and politely declined in under a minute. Without clear rules, that same call becomes a full intake for a job you were never going to take, and you pay for every minute of it.
Your escalation rules. Deciding early which calls go to a human keeps those calls short on the platform side. A call that spends four minutes trying to handle something it was never going to handle costs you four minutes and a frustrated customer.
Robocall and spam handling. These are cheap individually and add up. Any competent system should identify and end them quickly rather than running a full greeting on a dialer.
Your own opening. Long branded greetings sound professional and consume minutes on every single call, including the wrong numbers. A concise greeting that gets to the caller's problem is both better service and cheaper.
Two things that do not move your bill on a flat-plan structure are worth naming too, because owners worry about them. A long call that ends in a booked job is the best possible use of a minute and you should never optimize against it. And a busy season does not compound, because the overage rate is flat rather than escalating.
There is also a category of usage that is genuinely new rather than shifted. If you were previously letting the phone ring after five o'clock, the after-hours block is not minutes moving from a person to a system — it is calls that were being lost and are now being answered. Those minutes cost money and are the most profitable minutes on the bill, which is the whole argument of our after-hours playbook. Budget for them rather than being surprised by them, and measure them separately from your daytime volume when you size a tier.
Five questions that expose any quote
Ask these five, in writing, of every vendor including us.
- What is the total bill in a month where I take twice my normal call volume? One number. If they cannot produce it, their pricing is not forecastable and you should not sign it.
- What is the overage rate, is it flat, and is it published? A rate that escalates with usage is a metered plan wearing a subscription costume.
- List every fee that is not the monthly price. Setup, per-seat, per-booking, integration, telephony, number rental, recording, transcription.
- What is the contract term and what does it cost to leave? Month-to-month and annual are different products at the same price.
- What do I stop paying for elsewhere? Make them name the tools their platform replaces, then check that list against your own subscriptions.
Those five questions turn four incomparable quotes into four pairs of numbers. If a vendor resists answering any of them, that resistance is itself the answer.
The bottom line
The market clusters around $500 to $1,200 a month for a real AI answering platform in a service business, and within that band the structure matters more than the sticker. Flat monthly with an included minute pool and a published, flat, tier-declining overage rate is the model that lets an owner forecast, does not punish a good month, and does not give the vendor a reason to want your calls shorter.
Everything else — per-call fees, per-booking commissions, per-seat charges, setup fees, annual lock-in — is a way of moving cost off the headline and onto a line you did not price. Get the two numbers, normal month and worst month, for every vendor, subtract the tools each one replaces, and the decision usually makes itself.
Our full cost breakdown for the whole operations stack is in what AI operations actually cost, the current plan numbers are always on the pricing page, and if you want someone to size your call volume against the tiers before you commit, talk to a human.
