AI Cold Caller Costs, Law, and What It Can Actually Do

An AI cold caller is software that dials a list, talks with a synthetic voice, answers simple questions and tries to book a meeting with no person on the line. Published voice-agent rates at Retell AI run from $0.07 to $0.31 per minute (Retell AI pricing).

That minute price is the number that matters LEAST. Your meeting rate moves the cost per held meeting far more.

Then there is the law. In the US, an AI-generated voice used for telemarketing generally needs prior written consent. In the UK, an automated marketing call needs consent given specifically for automated calls, business numbers included. On a cold list, that ends the project before the demo does.

My bias, stated up front: my team books meetings through cold email and LinkedIn. We never staff a dialing floor or make pure cold calls, and phone only enters as an optional warm call to someone who has replied, engaged or opted in. If you want the meetings without the calling risk, see appointment setting.

Which products get sold as an AI cold caller?

Vendors sell four different things under this one name:

1. Autonomous voice agent. It places the call and does the talking. This is the AI cold caller proper, and the one this article prices.
2. AI-assisted dialer. A human rep talks; the software dials, skips dead lines and logs results.
3. Coaching tool. It listens to a human's call, prompts the rep and scores the conversation afterwards.
4. SDR support software. Research, notes, CRM updates and follow-up emails around the calls.

Only the first replaces the human voice, so the artificial-voice rules below hit it hardest. A good one can ask qualifying questions, handle scripted objections, hand off to a person, update your CRM and book a calendar slot.

Prospects do not care which platform you pay for. They care whether the contact was worth their time. We judge outbound by booked appointments, such as the 252 a UK B2B company got with us in 3 months. Hold a voice tool to a stricter test: count only the meetings that actually happen. For the human-versus-email choice, see cold calling or cold email.

What does the stack behind a voice agent need to include?

Under every demo sits the same stack. Vendors bundle the layers differently, and the bundle decides your bill.

Read pricing pages through this table. Bland AI says its per-minute rate covers the language model, speech-to-text and text-to-speech in one number, with telephony billed separately (Bland AI pricing). Retell bills each call to the nearest second, and when a call is transferred the agent fee stops while the telephony fee continues.

Capacity is written into the plans. Bland's self-serve Start plan costs $0.14 per minute with no platform fee and caps you at 100 calls a day. Its Build plan costs $0.12 per minute plus a $299 monthly platform fee, with 2,000 calls a day.

The trap is buying one layer and calling it a system. In outbound, I only see AI move revenue when the whole chain is built, from the list to the held meeting. A single automation glued onto a weak process changes nothing. I go deeper on when AI helps outbound and when it backfires.

Before you trust a demo, play a real prospect from your list, interrupt the agent mid-sentence, trigger a transfer and time it, then open your CRM and check every field it claims to write. Get the all-in minute cost in writing, telephony included.

How much does AI cold calling cost per held meeting?

Cost per held meeting takes three lines:

• Billed minutes = dials x average billed minutes per dial
• Held meetings = dials x meeting rate x show rate (the share of booked meetings that actually happen)
• Cost per held meeting = (billed minutes x price per minute + fixed monthly fees) / held meetings

Now a worked month. Take 2,000 dials, which is Bland's Start cap of 100 calls a day over about 20 working days. Assume 0.6 billed minutes per dial on average, because most dials are short or unanswered and a few run long.

Treat the funnel rates below as made-up inputs that show sensitivity. They are no vendor's benchmark and no result of ours, because our results come from email and LinkedIn, not the phone. Swap in your own first two weeks of data.

The minute price rose about 4.4 times from top to bottom. The cost per held meeting rose about 33 times. Meeting rate and show rate did almost all of that work.

Fixed costs shift it again. Move the mid scenario to Bland's Build plan and the minutes drop to $144, but the $299 platform fee lifts the month to $443, about $63 per held meeting across its 7 meetings, before list data, numbers, setup hours and the person who reviews calls every week.

Test small. If a funnel does not convert at 2,000 dials, I do not expect it to convert at 20,000. To compare routes, see what one B2B sales meeting costs and this AI SDR, human SDR or agency cost comparison. And the math only applies to numbers you may lawfully call.

What can you legally call in the US?

The federal rule is the FCC's Telephone Consumer Protection Act (TCPA) regulation, 47 CFR 64.1200. It restricts calls made with an autodialer or an artificial or prerecorded voice.

In its February 2024 Declaratory Ruling, the FCC confirmed that the artificial or prerecorded voice restrictions cover current AI technologies that generate human voices. The FCC consumer guide puts it plainly: AI-generated voice calls are illegal unless the consumer agreed to them or the caller is exempt. Prerecorded telemarketing calls must offer an opt-out at the start of the message.

For B2B sellers, the detail that matters is the wireless rule. The rule follows the phone number, and a business purpose earns no exemption. A founder's personal mobile is a wireless number, even when every call on it is about business.

The 8 a.m. to 9 p.m. window and the national do-not-call registry protect residential subscribers, and the same section extends both to wireless numbers.

What can you legally call in the UK?

UK law is stricter on automation. Regulation 19 of PECR allows direct marketing by an automated calling system only when the subscriber has already told the caller they consent. It defines an automated calling system as one that starts calls to more than one destination and transmits sounds that are not live speech, which describes a synthetic voice.

The ICO enforces this. It fined Home Improvement Marketing Ltd and Green Spark Energy Ltd a total of £550,000 for almost 12 million automated marketing calls in breach of PECR (Handley Gill, October 2025). Human agents triggered recorded clips with avatar software, and the calls still counted as automated.

The US risk is just as real. The FCC's February 2024 Declaratory Ruling holds AI-generated voices in calls to the same TCPA standards as artificial or prerecorded voices, and that includes prior express written consent for telemarketing. The FCC Enforcement Bureau then settled with Lingo Telecom for a $1M civil penalty and a compliance plan. Lingo had transmitted spoofed robocalls carrying AI-generated voice messages.

The ICO adds the working detail (ICO telephone marketing guidance). General marketing consent does not cover automated calls. B2B calls follow the same rules as calls to individuals, so check both the Telephone Preference Service (TPS) and the Corporate TPS (CTPS), and maintain your own suppression list.

I have not mapped phone rules country by country for the EU, so check each country's own rules before any automated call. My team treats Germany, Austria and the Czech Republic as prior-consent markets, so we do not cold email there without a compliant basis and use LinkedIn instead. This is not legal advice.

When is AI cold calling the wrong tool?

Use these rules before you book a single demo:

• Cold UK list? Automated voice is out. Use a live human on screened numbers, or written channels.
• Leads who opted in or replied? A callback from a voice agent or a human is the one place the phone fits.
• Cold US list of mobiles? No written consent means no AI voice.
• Weak list or vague offer? Nothing fixes that. In our campaigns, one message earned 2% replies at one company and 7.5% at another. A synthetic voice cannot close a gap like that, because it only changes how the words sound. Often you do not have a lead gen problem. You have a packaging problem.
• Buyers who check you out first? A call from nowhere gives them nothing to check. I put a LinkedIn touch before the email and plan five or six touches across different channels.
• Calls booked but no deals? Fix the sales process first.

Calls and deals are separate stages, and a voice agent only touches the first.

One of our B2B clients got 86 qualified leads and over $900k in new revenue in 2 months. Leads are the easy part to count. The money came out of the sales conversations that followed, and no voice agent runs those.

Which AI cold caller should you buy, or should you buy one at all?

On any cold list, every paid option shares one limit: it is legal only for numbers with the right consent, and that responsibility stays with the company running the campaign. After that, the choice is about caps and fixed fees. A self-serve per-minute plan suits a test, and a platform fee only pays off at volume.

A free tool labeled "AI cold caller" is often a simulator for practicing objections. It is useful for training reps, but it will not run outbound.

Fix the list, the offer and the follow-up first. A fresh campaign switched on inside a popular automation platform usually brings zero new clients.

Before any demo, answer yes or no:

1. Do I have consent that covers an AI voice for every number on this list?
2. Does my offer already book meetings through another channel?
3. Do I know my current meeting rate and show rate?
4. Can someone review calls and transfers every week?
5. Will the vendor's numbers land in my CRM as held meetings, the unit I will judge it on?

Two or more "no" answers? Stop. Fix the funnel first, then scale. And keep the phone for people who already know your name.

Before you pay for a voice agent, bring your list and your consent status to a 30-minute call and I will tell you what I would test first.