AI cold calling uses a synthetic voice agent to dial prospects, hold the opening conversation and book a meeting. Published 2026 platform rates run $0.05 to $0.31 per minute. Before you buy any of it: the FCC ruled in February 2024 that AI-generated voices count as artificial under the TCPA, which changes who you are legally allowed to call.
Most articles on this topic compare dialing speed. The thing that actually decides whether AI cold calling works for your business is not the tool, and not the per-minute rate. It is whether the people you want to call have given you permission to call them with a synthetic voice.
This page prices six platforms against their own published rate cards, read in September 2026, sets out the compliance position plainly, and is honest about which vendors will not tell you what they charge. We have been deploying conversational systems for Malaysian, Singapore and Hong Kong SMBs since 2019 and now deliver remotely for clients elsewhere.
The Legal Position Comes First
On 8 February 2024 the Federal Communications Commission unanimously adopted a Declaratory Ruling recognising that calls made with AI-generated voices are “artificial” under the Telephone Consumer Protection Act.
That single word does the work. The TCPA already restricted calls using an artificial or prerecorded voice, and the ruling confirms AI voices fall inside that restriction rather than outside it.
The practical consequences for anyone considering AI cold calling in the United States:
(1) Consumer cell phones are the hard case. Artificial-voice calls to consumers generally require prior express consent. An AI agent dialing a purchased consumer list is the scenario the ruling was written about.
(2) Business-to-business is a different risk profile, but not a free pass. Do-not-call obligations, state-level rules and carrier-level spam labelling all still apply, and none of them care that the caller was software.
(3) Disclosure is cheap insurance. Opening with a clear statement that the caller is an AI assistant costs a few seconds and removes the most damaging version of the complaint.
(4) Outside the US the rules differ but do not disappear. Malaysian and Singaporean deployments sit under the Personal Data Protection Act and national do-not-call registries, which govern how you obtained the number in the first place.
This is general information rather than legal advice. If you are dialing consumers at scale in a regulated market, have a lawyer read your call flow before it dials anyone.
What AI Cold Calling Actually Does
An AI cold calling system places outbound calls, speaks first, handles the opening exchange and either books a meeting or transfers a live human. Four components sit behind it.
(1) The dialer. Placing calls at volume, managing concurrency, and pacing so that carriers do not flag the number as spam.
(2) The voice agent. Speech recognition, a reasoning step, and synthesis fast enough that the prospect does not hang up during the pause.
(3) The transfer. Handing a warm prospect to a human mid-call, with context, rather than promising a callback.
(4) The write-back. Logging the outcome in the CRM, including the ones that went badly, which is the part most deployments skip.
The category splits cleanly on that first component. Platforms priced per minute give you a voice agent and expect you to bring the list and the strategy. Seat-based dialers give a human rep a faster phone and use AI to remove dead air between connects. They are not competing products, and buying the wrong one is the most common expensive mistake here.
How We Sourced These Numbers
Every rate below was read off the vendor’s own published pricing page in September 2026. Where a vendor publishes nothing, we opened and rendered the live pricing page to confirm that before saying so rather than inferring it. The legal position is quoted from the FCC’s own ruling page rather than from secondary coverage. We take no affiliate commission from any platform listed.
The 6 AI Cold Calling Platforms Worth Knowing
Per-minute voice platforms first, then the seat-based dialers, because the two are priced in units that cannot be compared directly.
1. Vapi
LOWEST PLATFORM FLOOR
Vapi charges for its own hosting only and passes speech, language and synthesis through at cost, dropping those to zero if you supply your own API keys.
- Hosting: $0.05 per minute
- Model costs: at cost, or $0 with your own keys
- Concurrency: 10 lines on Build, then $10 per line per month
- Compliance add-ons: HIPAA $2,000 a month, zero data retention $1,000 a month
2. Retell AI
CLEAREST COST BREAKDOWN
Retell AI publishes each component separately, which makes it the easiest platform here to model before committing.
- All-in range: $0.07 to $0.31 per minute
- Components: voice infrastructure $0.055/min, synthesis $0.015/min, language model $0.003 to $0.16/min, telephony about $0.015/min
- Free tier: $10 in credits and 20 free concurrent calls
- Concurrency: $8 per concurrency per month beyond the free 20
3. Bland AI
BEST FOR OUTBOUND VOLUME
Bland AI is built around outbound specifically, and is the only platform here publishing explicit daily and hourly call caps, which matters when you are dialing at volume.
- Start: $0.14 per minute talk, $0.05 per minute transfer, $0 platform fee, no card required
- Build: $0.12 per minute talk, $0.04 transfer, $299 a month platform fee
- Concurrency: 10 calls on Start, 50 on Build, custom on Enterprise
- Caps: 100 calls a day on Start, 2,000 a day and 1,000 an hour on Build
- Included in the rate: language model, real-time transcription and premium voices, with no token charges
The Build plan only makes sense above roughly 15,000 minutes a month, which is where the $299 fee is repaid by the $0.02 per minute saving.
4. Synthflow
ENTERPRISE PROCUREMENT
Synthflow has moved to enterprise-only and publishes a single number: contracts start at $30,000 annually.
- Entry point: $30,000 a year, published
- Priced on: call volume, concurrency, telephony setup, integrations, security requirements
- Included: launch support and security review
- Not published: per-minute rates
5. Nooks
HUMAN REPS, FASTER
Nooks is a parallel dialer for human reps rather than an autonomous agent. We rendered the live pricing page to check: it shows a contact form and no numbers.
- Pricing: not published, custom quote only
- Model: per seat, for human reps making live calls
- What the AI does: removes dead air, dials several numbers at once, connects the rep only when a human answers
6. Orum
TEAMS OF NINE OR MORE
Orum is the other seat-based dialer, and publishes one constraint that matters more than a price.
- Minimum: 9 seats
- Dials: no limit on any paid plan
- Free trial: limited to 500 dials
- Pricing: not published, request a quote
The nine-seat floor is the real filter. It rules out every team smaller than a full sales floor regardless of what the per-seat number turns out to be.
AI Cold Calling Platforms Compared
Every AI cold calling platform above, side by side, as published in September 2026.
| Platform | Published rate | Type | Concurrency | Best fit |
|---|---|---|---|---|
| Vapi | $0.05/min + models | Voice agent | 10 lines, +$10 each | Engineering teams |
| Retell AI | $0.07-$0.31/min | Voice agent | 20 free, then $8/mo | Transparent budgeting |
| Bland AI | $0.12-$0.14/min | Voice agent | 10 to 50 calls | Outbound volume |
| Synthflow | $30,000/yr | Voice agent | Custom | Enterprise procurement |
| Nooks | Not published | Seat dialer | Per rep | Existing rep teams |
| Orum | Not published | Seat dialer | Per rep, 9 min. | Sales floors of 9+ |
?Where did your call list come from?
Answer this before comparing a single price. If the numbers were given to you by people who asked to hear from you, every option on this page is open. If they were bought, scraped, or inferred, you are in the scenario the FCC ruling addresses and the tool you pick is the least of your problems. A list you cannot explain the origin of is not a cheaper list. It is an unpriced liability, and it is the reason most AI cold calling pilots get quietly shut down rather than scaled.
What Goes Wrong With AI Cold Calling in Production
Five failures account for most of the pilots we are asked to rescue.
(1) Carrier spam labelling. Dial enough numbers from one caller ID and carriers mark it “Scam Likely” before anyone picks up. Answer rates collapse and no amount of script tuning recovers it. Rotate numbers, register them properly, and watch answer rate by number rather than in aggregate.
(2) The opening three seconds. Cold calls are abandoned faster than any other call type. A voice agent that pauses before its first word loses the call before it has said anything. This is the one place where paying for a faster reasoning model is clearly worth it.
(3) Transfers that drop context. The agent qualifies a prospect, transfers to a rep, and the rep opens with a question the prospect already answered. That is worse than not transferring, because the prospect now knows they repeated themselves to a machine.
(4) No write-back on bad outcomes. Teams log the meetings and lose the refusals. Six weeks later nobody can tell whether the list is wrong, the script is wrong or the offer is wrong, because only the successes were recorded.
(5) Scaling before the script converts. An agent dialing 2,000 numbers a day with a script that does not work produces 2,000 negative brand impressions a day. Prove the conversation works at 50 calls before you touch the concurrency setting.
Buy a Platform, or Have the System Built
For most teams a platform above is the right answer. Two situations change that.
Buy a platform. Correct when the call flow is simple, the list is clean and consented, and your CRM is one the platform already connects to. Start on Retell’s free concurrency or Bland’s zero-platform-fee tier and prove the script before spending anything meaningful.
Have the system built. Correct when the call has to read from or write to systems with no ready connector, when it must switch language mid-call, or when a wrong statement on a recorded line carries a regulatory consequence.
The Crunch has been deploying conversational systems for Malaysian, Singapore and Hong Kong SMBs across healthcare, retail, property and education since 2019, with trilingual support across English, Bahasa Malaysia, Mandarin and Cantonese, and a 30-day deployment timeline for mid-tier scope. We deliver remotely for clients outside the region. Engagements start from USD 1,500 to USD 2,000.
We are worth a conversation when the call flow has to integrate with something awkward or operate across languages. Below that, a platform from the table above is cheaper and faster and you should take it. For the underlying technology see our guide to the AI voice agent stack; for inbound rather than outbound see AI customer service agents; for the wider sales picture see AI agents for sales; and for cost context see AI agent pricing. When you are ready to scope something, request a proposal.
01Is AI cold calling legal?+
It depends entirely on who you are calling and whether they consented. On 8 February 2024 the FCC adopted a Declaratory Ruling recognising that calls made with AI-generated voices are “artificial” under the Telephone Consumer Protection Act.
(1) The TCPA restricts calls using an artificial or prerecorded voice, and the ruling places AI voices inside that restriction.
(2) Artificial-voice calls to consumers generally require prior express consent.
(3) Business-to-business calling carries a different risk profile but do-not-call obligations and state rules still apply.
This is general information, not legal advice. Have a lawyer review your call flow before dialing at scale.
02How much does AI cold calling cost?+
Published platform rates run from $0.05 to $0.31 per minute in 2026.
(1) Vapi charges $0.05 a minute for hosting, with model costs at cost or zero if you bring your own keys.
(2) Retell AI publishes an all-in range of $0.07 to $0.31 a minute.
(3) Bland AI charges $0.14 a minute with no platform fee, or $0.12 with a $299 monthly fee.
(4) Synthflow starts at $30,000 a year on an enterprise contract.
Seat-based dialers such as Nooks and Orum publish no pricing at all and quote per rep instead.
03Does AI cold calling actually work?+
It works where cold calling already works, at lower cost per dial. It does not create interest that was not there.
The teams getting results treat it as a volume multiplier on a script and list that already convert with humans. The teams getting nothing use it to test whether cold calling works at all, and discover expensively that their offer was the problem rather than their capacity.
04Will prospects know it is an AI calling?+
Most will, within a few seconds, and the ones who do not usually work it out on the first unexpected question. Attempting to disguise it is the version that generates complaints.
Disclose in the opening line and offer an immediate route to a person. Beyond the reputational argument, disclosure is the cheapest available mitigation if a call is ever challenged.
05Why do my AI cold calls get marked as spam?+
Because carriers score the number, not the content. High volume from a single caller ID with low answer duration is the pattern that triggers a “Scam Likely” label.
(1) Register your numbers properly with the carriers rather than dialing from unregistered lines.
(2) Rotate across several numbers instead of pushing all volume through one.
(3) Track answer rate per number weekly, since a labelled number shows up as a sudden drop on one line while the others look fine.
06What is the difference between an AI voice agent and an AI dialer?+
A voice agent makes the call and does the talking, priced per minute. An AI dialer makes a human rep faster by dialing several numbers at once and connecting them only when a person answers, priced per seat.
They solve opposite problems. If your constraint is that you have no reps, you want a voice agent. If your constraint is that your reps spend half their day listening to ring tones, you want a dialer. Buying the wrong one is the most common expensive mistake in this category.
07How many calls can an AI agent make per day?+
Platform caps vary widely and are published less often than per-minute rates. Bland AI is the clearest, capping Start at 100 calls a day and Build at 2,000 a day and 1,000 an hour.
Concurrency is the practical limit rather than the daily cap. Retell gives 20 free concurrent calls, Vapi 10 lines on Build, Bland 10 on Start and 50 on Build. Concurrency is what determines whether you finish a list in a morning or a week.
08Can AI cold calling transfer to a human?+
Yes, and it is usually the highest-value thing the system does. Bland prices transfer time separately and cheaply at $0.04 to $0.05 a minute, which makes warm handoffs affordable enough to use as the default rather than the exception.
The thing to test before launch is whether context travels with the transfer. A rep who opens by asking what the prospect already told the agent turns a good handoff into a bad impression.
09Is AI cold calling allowed in Malaysia and Singapore?+
The FCC ruling is a United States instrument and does not apply directly, but the underlying obligations do not disappear.
Malaysian and Singaporean outbound sits under the Personal Data Protection Act and national do-not-call registries. The governing question is usually how you obtained the number and what the person consented to at that point, rather than whether the voice on the line is synthetic.
10How do I measure whether AI cold calling is working?+
Measure conversations, not dials. A system making 2,000 calls a day with nobody staying on the line is producing negative brand impressions at scale.
(1) Answer rate per number, weekly, to catch carrier labelling before it spreads.
(2) Median call duration, since cold calls that end in under ten seconds never had a chance.
(3) Transfer-to-meeting rate, which tells you whether the agent is qualifying or just passing everyone along.
(4) Complaints and opt-outs, tracked deliberately, because this is the metric that ends the programme if it is ignored.
