- An AI voice agent is a one-time build in the low thousands, then usage fees of pennies per call. A receptionist runs about $37,000 a year before taxes and benefits, for roughly 40 of the week's 168 hours.
- Platform minutes run about $0.08 to $0.16, so a typical four-minute intake call costs well under a dollar in usage, no matter what time it comes in.
- The build price is driven by intake complexity and integrations: a 13-step intake wired into your CRM, docs, and dispatch costs more than a simple after-hours answering flow.
- It's not worth it for everyone: if you get a handful of calls a week and someone reliably answers them, keep your setup and spend nothing.
The owner of a plumbing company used to answer the same intake questions himself, call after call, week after week. The voice agent I built him runs his exact 13-step intake on every call now, and it gave him back more than 20 hours a week. So what does something like that actually cost? A one-time build in the low thousands, then pennies per call to run. Here are the real numbers, line by line, next to what the alternatives cost.
How much does an AI voice agent cost?
Two numbers, honestly separated: the build and the running costs. The build is a flat, one-time fee in the low thousands, scoped to your intake and your tools. Running it is metered usage plus modest infrastructure, all billed to accounts you own:
| Line item | What it is | What it runs |
|---|---|---|
| The build | Your intake scripted, wired, and tested end to end | A few thousand dollars, one time |
| Platform minutes | The metered cost of the agent talking | ~$0.08 to $0.16 per minute |
| Telephony + infrastructure | Phone numbers, call routing, hosting | $150 to $1,200+ a month, usage-driven |
| AI model fees | The thinking behind the answers, metered | Pennies per call |
| Maintenance plan | Optional: monitoring, tuning, improvements | Separate, $300 to $2,500 a month |
Do the per-call math once and it stops being abstract: a typical four-minute intake call is about $0.32 to $0.64 in platform minutes plus pennies in model and telephony fees. Well under a dollar, at any hour, with no busy signal.
How does that compare to hiring someone?
The median receptionist wage is about $37,000 a year before taxes and benefits, and that buys roughly 40 hours of coverage a week, one call at a time. The honest comparison isn't "fire the human": it's what each option actually delivers for the money.
- Warm and human
- Handles true judgment calls
- ~40 hrs/week of coverage
- One call at a time
- Sick days, turnover, retraining
- Live voice after hours
- Takes a message, rarely books
- Reads from a generic script
- Doesn't know your business
- Costs climb with call volume
- 24/7, including holidays
- Books, quotes, logs to your CRM
- Concurrent calls, no busy signal
- Runs your exact intake script
- Escalates judgment calls to a human
The agent isn't warmer than a great receptionist, and it shouldn't make judgment calls. What it does is refuse to sleep, take two calls at once, and follow your intake perfectly on call four hundred, and it charges you cents for it.
What are you actually paying for? Coverage.
Your business week has 168 hours in it. An office phone gets answered for about 50 of them, and roughly 40% of appointments get booked outside business hours:
That grid is the real product. A dead water heater at 11pm is a same-day, high-margin job for whoever answers, and before the agent, that plumbing company's after-hours emergencies went to voicemail, which is to say, to a competitor.
What drives the build price up or down?
Complexity and connections. A simple after-hours answering flow sits at the bottom of the range. The price climbs with the depth of the intake and the number of systems the agent writes to. That plumbing build runs a 13-step intake, quotes from a knowledge base, logs the job to the CRM, generates a work order, posts a team notification, and texts the customer and the on-call tech. Every one of those wires is build time, and every one removes a manual step the owner used to do by hand.
A good build also ships with a trust loop, and you should ask for one. On that plumbing system, every job the agent books posts a notice to the team's Slack for a quick human audit, so the owner can check what the agent said and did in seconds instead of taking it on faith. You're not buying blind autonomy; you're buying work you can inspect.
Two cost-control rules I give every client. First, the infrastructure gets billed to accounts you own, itemized, so you're never renting your own phone system back from a vendor. Second, start with the one call flow that's bleeding the most, usually after-hours or overflow, and expand once it's proven. If a voice agent is just one piece of a bigger leak, the AI operating system cost breakdown covers how the full-system math works.
When is a voice agent not worth it?
When coverage is already solved. If you get a handful of calls a week and a human reliably answers them, keep your setup and spend nothing. The build pays for itself on volume and misses: calls that ring out during your busiest weeks, after-hours demand that books with whoever picks up, or an owner spending unpaid hours as the switchboard.
Here's the number to check before any of mine: how many calls did your business miss last month, and what's one booked job worth? If those two numbers multiply into more than a build costs, the phone is already deciding for you. If you want that math run on your actual call volume, that's the 15-minute call. I'll tell you what a build would cost for your intake, and if the honest answer is "keep your receptionist and add nothing," you'll hear that too.
