# AI sales calls: where they work and where they don't Source: https://dialmilo.com/hub/ai-sales-calls Updated: 2026-10-24 AI sales calls pay off on warm, consented lists and fail on bought ones. The law, a worked cost per sale, and a script structure you can run this week. AI sales calls work when the person already knows you and has agreed to hear from you: lapsed customers, renewals, people who asked for a quote and went quiet, upgrades to an existing plan. They fail on bought lists, both legally and commercially. An AI voice that discloses itself, checks it has the right person, presents one offer once and records a clear outcome can work through a warm list in a day for roughly 14–20¢ a talk minute. The script matters less than the list. Search for "AI sales calls" and you get two industries mixed together. Half the results are about software that sits on your human reps' calls, transcribing and coaching. The other half are about AI voice agents that make the call themselves. They solve different problems and they're regulated differently, so it's worth being clear which one you're buying. This article is about the second: the AI is the caller. It dials, introduces itself as an AI, holds the conversation, presents your offer and logs what happened. If you want the broader picture of outbound AI beyond sales, start with the [AI outbound calling guide](/hub/ai-outbound-calling). ## The honest map: where AI sales calls earn their keep The single biggest predictor of whether an AI sales call works isn't the voice, the model or the script. It's whether the person on the other end recognises why you're calling. Here's how common sales motions sort out: | Sales motion | Does an AI caller fit? | Why | | --- | --- | --- | | Following up a quote or enquiry that went quiet | Strong fit | They asked first. Speed matters more than polish. See [speed to lead](/hub/speed-to-lead). | | Renewal or upgrade for an existing customer | Strong fit | Fixed terms, a clear yes/no, a known relationship | | Win-back offer to a lapsed customer | Good fit | Relationship exists; the call doubles as research into why they left | | Booking a demo or appointment for a human closer | Good fit | AI qualifies and books; the human does the nuanced part | | Event or webinar confirmations with an upsell | Reasonable fit | Low-stakes, short, easy to measure | | Negotiated B2B deals | Weak fit | Pricing moves during the conversation; AI should not invent terms | | Cold calls to purchased consumer lists | Poor fit | Consent problems in the US and UK, and the lowest answer rates of any list | Notice the pattern. Everything in the top half has a reason for the call that the person will accept within the first ten seconds. Everything at the bottom relies on the caller's charm to overcome indifference, and charm is exactly what an AI disclosing itself as an AI is short of. [AI cold calling](/hub/ai-cold-calling) is possible in narrow B2B cases, but it's a different playbook. ## The law changes what "sales call" means for an AI Most guides to AI sales calls mention compliance in a paragraph near the end. It belongs at the start, because it decides which lists you can use at all. As of October 2026: **United States.** On 8 February 2024 the FCC adopted a declaratory ruling that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act. The [FCC's announcement](https://docs.fcc.gov/public/attachments/DOC-400393A1.pdf) is explicit that telemarketers need prior express written consent before robocalling someone, and that AI-generated voices are now held to the same standard. Separately, the FTC's [Telemarketing Sales Rule guidance](https://www.ftc.gov/business-guidance/resources/complying-telemarketing-sales-rule) limits calls to between 8am and 9pm local time and requires you to scrub lists against the National Do Not Call Registry at least every 31 days. That registry is not a niche list: the [FTC's 2025 Data Book](https://www.ftc.gov/system/files/ftc_gov/pdf/DNC-Data-Book-2025.pdf) counts about 259 million active registrations. **United Kingdom.** The ICO's [telephone marketing guidance](https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guide-to-pecr/electronic-and-telephone-marketing/telephone-marketing/) distinguishes live calls (screen against the TPS) from automated calls, where consent "must specifically cover automated calls". General marketing consent isn't enough. The ICO notes that guidance is under review following the Data (Use and Access) Act, so check the current version before a campaign. **European Union.** Article 50 of the [EU AI Act](https://artificialintelligenceact.eu/article/50/) requires that people are informed when they're interacting with an AI system, applying from 2 August 2026. On top of that sit each country's own telemarketing rules and GDPR. This isn't legal advice. It's a summary of published guidance as of October 2026, and the rules depend on who you call, from where, and what you sell. For the longer version see [is it legal to make AI calls](/hub/ai-phone-call-laws), and have your own counsel sign off the consent wording on your sign-up forms. The practical upshot is simple and, frankly, good for results: AI sales calls belong on lists of people who opted in to calls from you. Those are also the people most likely to answer and to say yes. ## Anatomy of an AI sales call that converts A good AI sales call is short. On a warm list, two to three minutes of talk time is typical for a call that ends in a decision. Five beats, in order: 1. **Disclosure and identity.** "Hi, this is Milo, an AI assistant calling for Northfield Fitness. Am I speaking with Sam?" Disclosure is on by default in Dial Milo. If the person says it's the wrong number, the call ends without the offer. 2. **The reason, tied to them.** "You were a member until March and I wanted to check in." One sentence. No small talk about the weather. 3. **One offer, stated once.** "We can bring you back at your old rate of $29 a month for the next six months." Milo presents the offer you wrote, word for word on the terms, and never invents or changes them. 4. **Questions answered from your terms.** Cancellation policy, start date, what's included. If a condition applies, Milo gives it when asked rather than reading out the small print unprompted. 5. **A recorded outcome.** Accepted, declined, wants a callback at a specific time, opted out, wrong person, no decision. Each one triggers a different next step. Two details from a real pilot we ran in October 2026 apply to any sales call. A "yes" spoken mid-sentence ("yes, but what about…") must not restart the pitch; the AI has to hear the question that follows. And the pause after someone picks up needs to stay under a second, because a silent line reads as a robocall and people hang up into it. > **Milo:** Hi, this is Milo, an AI assistant calling for Northfield Fitness. Is that Sam? > > **Sam:** Yeah, speaking. > > **Milo:** Thanks, Sam. You were a member with us until March, and we're offering past members their old rate of $29 a month for six months if they come back before the end of November. Would that be useful? > > **Sam:** Maybe. Is there a joining fee again? > > **Milo:** No joining fee for returning members, and you can cancel with a month's notice. > > **Sam:** Fine, go on then. > > **Milo:** Great. I've noted that you'd like to rejoin at $29 a month. The team will send you the details by text and email today. That call is illustrative, and the business is fictional, but the shape is what you should aim for: under two minutes, one offer, a clean outcome. The "accepted" result is what triggers your own follow-up, whether that's a CRM task or an automated text and email from your marketing tool. ## What AI sales calls cost, worked through Dial Milo is pay as you go with no subscription. An outbound call has two parts: the phone line, priced for the exact number you dial and the caller ID you call from and billed per started minute, and the AI at 9.57¢ per talk minute. US landlines and mobiles start from 4¢ a minute on the line; a UK mobile from a UK number is 8.5¢. That puts an answered call at roughly 14–20¢ a minute all in for US and UK numbers. You can price any specific number on the [rates page](/rates). **Worked example (illustrative figures).** Say you have 1,000 US customers who lapsed in the last year and opted in to calls. - Assume 30% answer on the first round: 300 conversations. - Assume an average of 2.5 minutes of talk each. The line bills 3 started minutes per call: 300 × 3 × 4¢ = **$36**. - AI talk time: 300 × 2.5 × 9.57¢ = **$71.78**. - The 700 calls that ring out or hit voicemail cost little. Calls that don't connect have no minute charge; voicemail is detected and Milo hangs up without leaving a message. Allow, say, **$15** for those short connections. - Total for the round: about **$123**. If 6% of the people who answer accept the offer, that's 18 sales, or roughly **$6.80 per sale**. Change any assumption and the arithmetic changes with it, which is why the [spend calculator](/help/using-the-spend-calculator) asks for your own answer rate and call length before it estimates anything. For a real-world anchor: in an October 2026 pilot with an online casino operator, 121 dial attempts across 61 people cost $11.53 in total, about 9.5¢ per attempt, because most attempts were unanswered or voicemail. Cost per attempt on a real list is usually far below the per-minute rate. ## Script testing beats script writing Most teams spend a week polishing one script. A better use of that week is to write two and let the calls decide. Dial Milo runs up to five scripts side by side in one campaign; each person keeps the script they first heard, and the comparison only declares a winner when it's at least 95% sure the difference is real. The [A/B testing guide](/help/ab-testing-scripts) covers the mechanics. Good things to test, one at a time: - **The reason line.** "We noticed you haven't been in since March" against "We're inviting past members back". - **Price first or benefit first.** Lead with the $29, or lead with what's new. - **Offer size.** A smaller offer accepted more often can beat a bigger one, once you count cost per sale. Test the opening, not the whole script. The first fifteen seconds decide whether there is a conversation at all, so that's where a difference shows up fastest. ## Launching your first AI sales campaign Lapsed customers, quiet quotes, upcoming renewals. Confirm they opted in to calls. Remove anyone who asked not to be contacted; Dial Milo's do-not-call list does this automatically once someone is on it. Price, duration, conditions, deadline. Milo presents these and nothing else, so if a term isn't written down, it won't be promised. Calling hours in each person's local time (the default is 10:00 to 20:00), two attempts a few hours apart, a contact cooldown so nobody hears from you twice in a week, and a daily budget cap. Use the browser demo or a test call. Answer as a sceptical customer would: interrupt, ask about cancellation, say "not now". Read the transcripts of the first hundred calls before you load the next thousand. The AI campaign report labels each person with what they need next, so you can see where the script loses people. ## What to measure Ignore call volume; it's the easiest number to inflate and the least useful. Track these instead: - **Answer rate** by time of day and day of week. - **Acceptance rate per answered call**, per script. - **Cost per acceptance**, from the campaign's own spend, not an estimate. - **Opt-out rate.** If more than a small fraction of answered calls end in "don't call me again", the list or the reason line is wrong. - **Callbacks.** People ring back a number that just called them. Dial Milo recognises them, thanks them for calling back, confirms who they are and continues the same conversation, so those calls count toward the campaign. When those numbers look healthy on a warm list, you've found the shape of a sales call worth repeating. When they don't, change the list before you change the voice. You can [sign up](/signup) and try the browser demo before you put any money in. ## Frequently asked **What is an AI sales call?** The phrase covers two different products. One is software that listens to your human reps' calls and coaches them or writes notes. The other is an AI voice agent that places the call itself, holds the conversation, presents an offer and records the outcome. This article is about the second kind: the AI is the caller, and it says so at the start of the call. **Are AI sales calls legal?** They can be, but the rules are stricter than for a human rep. In the US the FCC ruled in February 2024 that AI-generated voices count as artificial voices under the TCPA, so marketing calls using them need the person's prior express written consent. In the UK, ICO guidance says automated marketing calls need consent that specifically covers automated calls. In practice that means AI sales calls belong on lists of people who opted in, not on purchased data. This isn't legal advice; check your own position. **How much does an AI sales call cost?** With pay-as-you-go pricing you pay for the phone line and the AI's talk time. On Dial Milo, US line rates start from 4 cents a minute and the AI is 9.57 cents per talk minute, so an answered call to a US or UK number runs roughly 14 to 20 cents a minute all in. Calls that don't connect have no minute charge, which is why the average cost per dial attempt on a real list is usually well under the per-minute rate. **Can an AI close a sale on the phone?** For simple offers, yes: a renewal at a stated price, a booked appointment, an upgrade with fixed terms, a time-limited discount. The AI presents the offer once, answers questions from the terms you wrote, and records an acceptance. For negotiated or high-consideration deals it works better as the qualifier that books the human closer. **Will people hang up when they hear it's an AI?** Some will, and you should let them. Disclosure is required or expected in most places, and hiding it destroys trust the moment someone notices. The calls that work are short, relevant to something the person already did with you, and give them an easy way to say no or ask to never be called again.