Two decks land in your inbox the same week. Both say AI, both say outbound, both quote a per-minute rate. One product is called an AI SDR, the other is called AI cold calling. The demo recordings sound nearly identical, so you compare the rates and shortlist the cheaper one.
The two products do different jobs. An AI SDR works a list you already have a relationship with, or a documented basis to contact: inbound leads, trial signups, lapsed customers, people who left a badge scan at your stand. Its job is qualification and booking. AI cold calling dials people who have never heard of your company, and from the consent question down to the opening of the script, it is a different build.
Buying one while expecting the other is an easy way for a project like this to disappoint. Nothing technical has to break for that to happen. The agent can hold the conversation and hit its latency budget, and still be pointed at the wrong kind of list.
The two motions people conflate: AI SDR vs AI cold calling
Underneath, the stack is the same. Same speech recognition, same voice synthesis, same model, same reply latency. Nothing in the technology decides which motion you are running. The list decides that, and the list then decides almost everything else.
- Warm motion, sold as an AI SDR. The person has some prior connection to you: a form, a trial, a purchase, a callback request. You can point to the record, the opening line references it, and the job is to establish whether there is real need and book it.
- Cold motion, sold as AI cold calling. The person has no connection to you at all. The opening has to earn the next moment from nothing, you need a defensible answer to why you may call this number, and somewhere for the refusals to go.
The technology does not decide which motion you are running. The list does.
What an AI SDR actually does
Strip the label off and an AI SDR has a narrow, checkable job: on a list you already own, work out who is worth a salesperson's time, and book it.
Work the list and detect voicemail, so the agent drops off instead of talking to a machine. Ask each vendor what their build does about retries and local calling hours, and get the answer in writing.
Check you have the right person, reference why you hold their number, and work through the qualifying questions your sales team actually uses. The agent handles being interrupted, so someone who cuts in gets an answer instead of the rest of a scripted paragraph.
Qualification is only worth something if it leaves a record: need, timing, decision role, objections raised and how they were handled. Agree with the vendor which fields you get. A booked meeting with an empty record is not a qualified lead.
Decide before you start what happens to a qualified contact, and get the vendor to write down which handover they actually support.
Replies land in about 0.3 seconds. Warm contacts are not surprised to hear from you, but they will still hang up on a gap that feels like a bad line.
The metric here is qualified meetings, and what you actually know about each one afterwards.
What changes the moment you dial a stranger
Cold outbound is a separate product because four things move at once, none of them about the voice.
- The consent basis. Whether you may call a number, and on what grounds, varies by country and by how the number was obtained. So do the rules on telling people they are speaking to an AI. Check both for every market before the first call, not after the first complaint.
- List hygiene. Where the list came from, whether that provenance is documented, and whether a do-not-call request captured on the call actually suppresses the number next time. On a warm list this is housekeeping. On a cold list it is most of the risk profile.
- Number reputation. Carriers filter calls, and business numbers can be flagged or blocked. Ask how caller IDs are separated between your warm and cold motions before you start.
- The script. A cold opening has to say who is calling and why immediately, and needs a clean exit the moment the answer is no. A warm script can afford to be conversational, because the context is already shared.
Who owns the compliance obligation
We build the agent and run it on our own servers. The lawful basis for calling a specific list in a specific country is a question for the buyer and their counsel; no vendor's architecture answers it, and this is not legal advice. A vendor who waves the question away is selling you a problem with a per-minute price attached.
One part of it we can take off your plate: speech recognition, synthesis and the language model all run on our own servers, so call data stays in-region and does not depend on third-party APIs. That answers the data question, not the consent question.
Connect rate is not conversion
Connect rate is the share of dials that become a live conversation. Whether yours differs between a warm list and a cold one is something to measure on your own numbers, not to take from a vendor's slide. It gets quoted because it is easy to count. It is not what decides whether the project pays.
We bill talk time only, with no subscription, so a dial that rings out or hits an answering machine costs nothing in agent minutes. That is useful, and it also hides the difference between the two motions inside the line item you are comparing. Price the outcome instead.
Bene Qualifier is 0.25 per talk minute, excluding VAT. Assume a three-minute warm qualifying conversation and substitute your own average: that conversation costs 0.75. Assume, too, that one conversation in eight ends in a qualified, booked meeting, and the talk time costs 6.00 per meeting. Now the cold version. Assume, again only for the example, one-minute conversations at the same rate. If it takes forty of those to reach one qualified, booked meeting, the talk time costs 10.00 per meeting.
Every ratio and duration above is an assumption for the sake of the sum, so substitute yours. What matters is the shape: the per-minute rate barely moved, and the number of strangers you had to touch moved by a lot. That last number is where the telephony bill, the caller ID reputation and the compliance work sit, and none of it shows up on a per-minute quote.
Connect rate tells you how many numbers you have to burn. Conversion tells you what an outcome costs.
It is also why the commercial model we are moving toward is pay-for-results: per sale, per qualified lead, per desired action. Once the price is attached to the outcome, the argument about which motion is cheaper per minute stops mattering.
Which Benerra agent fits which motion
We sell four things, and none of them is a cold-calling bot. What varies between them is the job, not the standing of your list.
Rates exclude VAT, and the same nominal figure applies in EUR and USD as regional parity rather than a currency conversion. Latin America is priced at roughly half those rates. Volume tiers run from 25,000 minutes a month to a million, discounting the agents down to minus 25% and Locator to minus 62.5%. A trial is priced 20% above base.
Locator is the analytics layer that runs before any agent, and it is where our rollout order starts: speech analytics first, then the inbound agent, then Qualifier, then Bene Sales. Running Locator first tells you what your own team already says on the calls you are thinking of handing over. For cold outbound, Qualifier or Bene Sales will dial the list at the same rates, with a script built for strangers and compliance work that has to be done for each market.
The questions to ask before you buy
Take these to every vendor on your shortlist, us included. The answers separate the two products faster than any demo recording.
- Which list is this quote priced against? Ask them to write down whether the pricing assumes contacts who opted in or a purchased list of strangers.
- Who owns the lawful basis for calling? If the answer is a shrug or a slide that says fully compliant, the answer is you.
- What does your build do about telling people they are speaking to an AI, and what will you put in writing about it?
- Where do refusals go? An opt-out captured on the call and never written back to a suppression list is not an opt-out.
- What exactly am I billed for: talk time, dials, connected calls, seats, a monthly platform fee?
- What is the outcome metric, and is it in the contract? Qualified meetings booked is a metric. Minutes delivered is an invoice.
- Where does the call data sit, and which third-party services see it on the way?
The difference between an AI SDR and AI cold calling is not a feature list. It is which list the agent is allowed to dial.
Price both motions on your own numbers
Put your call volume, your region and the product you are pricing into the cost calculator and it will price the talk time itself. Then finish the sum by hand: multiply the cost of one conversation by the number of conversations it takes you to book one meeting, once on your warm numbers and once on your cold ones. The per-minute rates sit in the pricing block on the homepage, so there is nothing to unlock and nobody to email first.
If the two sums land close together, you are pricing the per-minute line rather than the cost of reaching the conversation. That gap is the whole difference between the two products.