Loading
AI made outbound calling cheap, which is exactly what makes it risky. With 86% of consumers ignoring unknown numbers, the win is not more dials. It is calling a short, warm list within minutes, with documented consent. Here is what to automate, what to skip, and what the TCPA actually requires in 2026.
Last updated: August 2026. Hiya's State of the Call 2026 report, built on a Censuswide survey of more than 12,000 consumers across six countries, found that 86% of people do not answer calls from numbers they do not recognize. That one number should change how you think about AI outbound calling. Dialing is effectively free now, which tempts owners to dial more. The evidence points the other way: the money is in calling the small number of people who are already expecting to hear from you, fast, and leaving everyone else alone. This guide covers which outbound calls are worth automating, which ones to skip, what US law requires before an AI voice dials a phone, and what the whole thing costs.
Key takeaways
- 86% of consumers do not answer unknown numbers, and 20% said no piece of information would make them more likely to answer a business call (Hiya, State of the Call 2026). Volume dialing is the wrong instinct.
- A callback is not a cold call. Returning contact to someone who just filled out your form is the highest-value, lowest-risk outbound call you can automate. Cold-dialing a purchased list is the opposite on both counts.
- The FCC ruled in February 2024 that AI-generated voices are "artificial" under the TCPA, so an outbound AI call needs the called party's prior express consent (FCC Declaratory Ruling, February 8, 2024). Statutory damages start at $500 per call.
- The consent rules are in flux. The Eleventh Circuit vacated the FCC's one-to-one consent rule in January 2025, and in February 2026 the Fifth Circuit held the TCPA requires only prior express consent, not written consent. Build to the strictest rule that applies to you, not the most convenient one.
- 73% of consumers say businesses should identify themselves on caller ID, and a verified business name is the single piece of information most likely to earn a pickup (Hiya, 2026). Getting your number recognized is worth as much attention as the script.
AI outbound calling is software that places a phone call on your behalf and holds the conversation itself, rather than dialing a number and handing it to a person. Something triggers the call. A form submission, a call you missed, an appointment 24 hours out, a quote that went quiet. The system then dials, speaks, listens, and acts on what it hears: qualifying the lead, booking a slot on your calendar, confirming or rescheduling an appointment, taking a message, or transferring a live caller to whoever is on duty.
It differs from a robodialer in that nothing is pre-recorded and nothing follows a fixed menu tree. The person you called can interrupt, ask a question, or change the subject, and the agent responds. AIEmply's agents connect and respond in under 3 seconds, run 24/7/365, handle unlimited simultaneous calls, and speak 50+ languages. Automated lead follow-up calls, follow-up campaigns and lead re-engagement are included from our Growth plan upward.
Mostly no, and any vendor who skips this part is selling you something. Hiya's State of the Call 2026 report, based on Censuswide research with more than 12,000 consumers in the US, Canada, UK, France, Germany and Spain between December 17, 2025 and January 2, 2026, found that only 14% of people answer immediately when an unrecognized number rings. Of the rest, 41% ignore it to see if a voicemail lands and 28% reject the call outright. Respondents reported an average of 7 unwanted calls a week, and 34% said their trust in phone calls had fallen over the previous 12 months.
Worth naming the obvious: Hiya sells call-protection software, so this is a vendor survey about a problem that vendor solves. The methodology is disclosed and the sample is large, which is more than most numbers on this topic can claim, but read it with that in mind.
Either way, that is the ceiling any outbound program runs into, human or AI. What moves it is being recognized and being expected. The same survey found 73% of consumers agree businesses should identify themselves on caller ID. Asked which single piece of information would most make them answer a business call, 29% chose a verified business name and 24% chose seeing the number had not been flagged as spam by their carrier. Another 20% said none of the options would help.
Asked what would make them most likely to answer a business call, 20% of consumers chose "nothing."
Hiya, State of the Call 2026 (Censuswide, 12,000+ consumers across six countries)
This is the lever most owners never touch, and it sits upstream of anything your script does. Four steps, in order of effort:
Sort every outbound call by a single question: does this person expect to hear from you? The closer the answer sits to yes, the higher the pickup rate and the lower your legal exposure. Both move together, which makes the ranking unusually clean.
In descending order of value, the list runs: new-lead callbacks, missed-call returns and appointment confirmations at the top, quote follow-ups in the middle, dormant-database reactivation below that, and cold outreach to a purchased list at the bottom, where we would tell you not to go at all. The first three are where nearly all the return lives for a service business, because they are calls the recipient is waiting for. That is exactly the condition the Hiya data says you need. Appointment setting rides along with all three: the same agent that qualifies the lead can book the slot before the call ends.
| Outbound call type | Who you are calling | Expected pickup | Consent risk | Automate it? |
|---|---|---|---|---|
| New-lead callback (web form, ad, chat) | Someone who contacted you minutes ago | High | Low | Yes, first priority |
| Missed-call return | Someone whose call you just missed | High | Low | Yes |
| Appointment reminder or confirmation | An existing customer with a booking | High | Low | Yes |
| Quote or estimate follow-up | Someone who received a bid from you | Medium | Low to moderate | Yes, with a cap on attempts |
| Dormant database reactivation | Someone who inquired months or years ago | Low to medium | Moderate to high | Only with documented consent |
| Cold outreach to a bought list | A stranger who never contacted you | Very low | High | No |
The bottom row is where the lawsuits are, and it is not a use case we support. If a missed-call return is your first priority, we compared the voice and SMS versions of it in our breakdown of missed call text back.
A callback returns contact to someone who reached out to you first. A cold call initiates contact with someone who did not. That distinction is not a marketing nicety. It drives your pickup rate, the tone of the conversation and your legal position all at once.
Picture a homeowner submitting a form on your site at 9pm, and your AI calling back at 9:01pm. The caller ID is expected. The topic is the one they raised sixty seconds ago. They handed you the number for this exact purpose. Dial a list you purchased and none of that holds, while every protection the TCPA and the Telemarketing Sales Rule provide is pointed at you.
Here is the rule we give clients. Automate outbound only to people who have contacted you, bought from you or booked with you, and only about the thing they came to you for. It keeps you inside the safest reading of the consent rules in this guide, and it is where the revenue sits anyway. For the underlying research on why callback speed matters this much, see Speed to Lead in 2026.
Yes, with consent. On February 8, 2024 the FCC issued a Declaratory Ruling confirming that calls using AI-generated voices are "artificial" under the Telephone Consumer Protection Act, which means they carry the same restrictions as any prerecorded or artificial-voice call and require the called party's prior express consent (FCC 24-17). It took effect immediately.
Then-Chairwoman Jessica Rosenworcel framed the intent plainly in the FCC's announcement:
"Bad actors are using AI-generated voices in unsolicited robocalls to extort vulnerable family members, imitate celebrities, and misinform voters. We're putting the fraudsters behind these robocalls on notice."
Jessica Rosenworcel, then FCC Chairwoman, February 8, 2024
The exposure is real. Under 47 U.S.C. § 227(b)(3), statutory damages are $500 per violating call, and a court has discretion to increase that award to as much as three times the amount, or $1,500, where it finds the violation was willful or knowing. There is no aggregate cap and each call counts separately, which is why private class actions tend to be the bigger practical risk for a small business than FCC enforcement. Note that this rule governs outbound calls. Answering your own inbound phone with an AI raises a different legal question, covered in Is an AI Receptionist Legal?
Enough that you can prove it in writing, because the standard itself is unsettled right now. FCC rules have long required prior express written consent for artificial-voice telemarketing calls to a mobile number, and mere prior express consent, which can be oral, for informational and transactional calls such as an appointment confirmation.
Two rulings have shaken that. In Insurance Marketing Coalition Ltd. v. FCC, No. 24-10277, the Eleventh Circuit vacated the FCC's one-to-one consent rule on January 24, 2025, holding the agency had exceeded its statutory authority. The FCC formally removed the rule in September 2025. Then in Bradford v. Sovereign Pest Control of TX, Inc., No. 24-20379, decided February 25, 2026, the Fifth Circuit found the statute asks for less than the FCC's regulation does. Chief Judge Jennifer Walker Elrod wrote that the court affirmed "because the TCPA requires no more for any type of pre-recorded calls" than prior express consent, which may be given orally or in writing.
That ruling binds Texas, Louisiana and Mississippi only, and it conflicts with the FCC's own regulation elsewhere. Do not read it as permission to loosen your process. Collect written, timestamped consent tied to the specific number. Store it where you can retrieve it. Honor a withdrawal the moment it is spoken. Build to the strictest rule that could apply to you and the circuit split becomes somebody else's problem.
Consent is the headline. Five other requirements catch service businesses far more often, and none of them are hard to comply with. All of them are easy to forget once calling is automated and nobody is watching each dial.
| Requirement | What it means in practice | Source |
|---|---|---|
| Calling hours | No telemarketing calls before 8am or after 9pm in the called party's local time, not yours | 47 C.F.R. § 64.1200(c)(1) |
| National Do Not Call Registry | Scrub against it unless an exemption applies | FTC Telemarketing Sales Rule, 16 C.F.R. Part 310 |
| Established business relationship | 18 months from the last transaction, or 3 months from an inquiry or application | 47 C.F.R. § 64.1200(f)(5) |
| Company-specific do-not-call list | If someone asks you not to call, you stop, even with an existing relationship | FTC Telemarketing Sales Rule, 16 C.F.R. § 310.4(b) |
| State AI-voice disclosure | California requires an automated call delivering a prerecorded message to disclose when the message uses an AI-generated artificial voice | Cal. AB 2905 (2024), amending Pub. Util. Code § 2874 |
That three-month inquiry window is the one that trips up reactivation campaigns. A lead who filled out your form 14 months ago and never bought falls outside it, so if their number sits on the Registry you need actual consent to call them. We treat dormant-database reactivation as a consent question first and a marketing question second for exactly that reason. Rules vary by state and this is general information rather than legal advice, so run your specific program past your own counsel.
Fast enough that you are the first business to call, which in practice means minutes rather than hours. The most-cited research here is the Lead Response Management study led by Dr. James Oldroyd (then a faculty fellow at MIT Sloan) with InsideSales.com in 2007, covering more than 15,000 leads and 100,000 call attempts. Calling within 5 minutes rather than 30 made a business roughly 100× more likely to reach a lead and 21× more likely to qualify one. Note the verbs. That study measured contact and qualification, and stated outright that it "did not address close ratios." Anyone quoting it as a conversion multiplier is overselling it.
Reality sits a long way from that. Harvard Business Review's 2011 audit of 2,241 US companies found 23% never responded to a web inquiry at all, and among those replying within 30 days the average was 42 hours. An always-on agent closes that gap without anyone sitting on call at 11pm. We go deeper into the evidence, including which speed-to-lead statistics are not trustworthy, in our speed-to-lead breakdown.
Getting this backwards is the most expensive mistake in the category. Inbound is where the intent already is. Invoca's Lead Conversion Benchmarks Report 2026, drawn from more than 70 million phone calls, found only 56% of callers to businesses reach a person, rising to 71% on calls lasting longer than 30 seconds. Invoca sells call tracking, so treat it as vendor research, though the sample is unusually large.
Do the arithmetic. Roughly 44% of people dialing you on purpose never get through.
Fix that before you spend a minute on outbound. A caller who picked up the phone and chose your business outranks any list you could dial, and they are already on the line. Our missed-call cost breakdown works through what that leakage is worth by industry, and what the data says about customers accepting an AI covers the objection most owners raise next. Outbound is the second layer, catching the form fills, the after-hours inquiries, the no-shows and the quotes that went quiet. Both layers run on the same agent, which is why we do not price them as separate products.
Priced two ways across the market. Platform tools that you configure yourself bill per connected minute, so the sticker looks cheap until you add your own build and maintenance time. Managed services bundle configuration, tuning and support into a flat monthly fee, which is where most service businesses land, because nobody on the team wants to own a call flow.
Either way, the number that matters is cost per booked job, not cost per minute. Run it against your own averages before you buy. Recover one $1,200 job a month and a flat monthly plan clears its own cost several times over. If your average ticket is $150 and you recover two, the math is tight, and you should know that going in.
AIEmply prices outbound as part of the same AI Employee that answers your phone, not as a separate seat or a per-dial fee. The Growth AI Employee is $399/month ($3,348/year) and is the first tier that includes full outbound calling, follow-up campaigns, lead re-engagement and Meta ad-form follow-ups, alongside an AI website chatbot and 250 included minutes at $0.25 per additional minute. Scale runs $599/month ($5,040/year) with 400 minutes at $0.20 and high-volume inbound and outbound. Starter, at $149/month, is inbound only.
Yearly billing is 30% off. No credit card is required to start, there are no setup fees, and you can cancel anytime. The guarantee is worth reading twice: "Billing starts only after your AI Employee is live. If the first month delivers no measurable result, the next month is free." Setup runs 1–2 weeks in most cases, from a 15-minute consult to 4–7 days of configuration and 3–5 days of testing.
Several places, and we would rather say so than pretend otherwise. Never automate the outbound call that delivers bad news: a job being rescheduled at short notice, a repair that came in over estimate, a complaint that already escalated once. Those calls exist to repair trust. An AI voice on the other end reads as a company that could not be bothered.
High-value negotiation belongs to people too. If you are talking through a $40,000 remodel bid, or a retainer with a client comparing three firms, that call is the sale. Automate the scheduling around it and leave the conversation alone.
Sensitive and emotional situations are the third category, which is why our agents are built to escalate rather than push through. Instant warm handoff to a live person is a standard feature, not an upgrade. The honest framing: AI outbound handles the calls nobody had time to make, and humans keep the ones that decide whether a customer stays.
We treat outbound as a follow-up engine attached to an inbound agent, not as a dialer. The same AI Employee that answers your phone in under 3 seconds calls a new web lead back, chases the estimate that went quiet, confirms tomorrow's appointments and re-engages a lead who went dark, then writes what it learned into your CRM.
It works with the tools you already run, including GoHighLevel, Salesforce, HubSpot, Zendesk and Pipedrive, industry systems such as Jobber, Housecall Pro and ServiceTitan via booking portals, real-estate stacks including kvCore, Follow Up Boss and Chime, plus Google Calendar and Outlook, and many more through custom API integrations.
On the parts specific to outbound: we configure calling windows to the recipient's time zone, cap attempts per contact, honor opt-outs the moment they are spoken, and disclose that the caller is an AI assistant where the law or the situation calls for it. We do not run cold campaigns against purchased lists. On Growth and above, a real person on our team reviews and refines your agent every week, so call #10,000 is better than call #1. Want to hear the behavior before deciding? Our live demo plays a real call, and how it works walks through setup.
AI made outbound calling cheap, and cheap is what makes it dangerous. With 86% of consumers ignoring numbers they do not recognize (Hiya, 2026), the businesses winning at outbound are not the ones dialing most. They are the ones calling a short, warm list within minutes, from a number people recognize, with documented consent behind every dial.
Get the new-lead callback right first. Add missed-call returns and appointment confirmations. Leave the cold list alone.
If you want to see what that looks like against your own call volume and job value, we will map it with you. 100% Answer Rate • Ready in 1–2 Weeks • Performance Guarantee. Book a 15-minute consultation and we will walk through your current follow-up gaps, what an AI Employee would pick up, and what it would cost. You can also compare plans or browse the setup for HVAC and plumbing, contractors, real estate and dental clinics.
Yes, with consent. The FCC confirmed in February 2024 that AI-generated voices count as "artificial" under the TCPA, so an outbound AI call requires the called party's prior express consent. You also need to respect calling hours of 8am to 9pm in the recipient's local time, scrub against the National Do Not Call Registry unless an exemption applies, and honor opt-outs immediately. Statutory damages start at $500 per violating call.
A robocall plays a fixed recording at you. An AI outbound call is a live two-way conversation: the agent listens, answers questions, handles interruptions, books a time on your calendar and transfers to a person when it should. Legally, though, the TCPA treats AI-generated voices the same way it treats prerecorded ones, so the consent requirements are identical. The technology differs; the compliance obligations do not.
Yes, and it is the single highest-value outbound job. A form submission triggers the call within seconds, the agent qualifies the inquiry, books an appointment if the fit is there, and writes the outcome into your CRM. Because the person contacted you moments earlier and supplied the number for that purpose, this is the lowest-risk outbound call you can automate and the one most likely to be answered.
Yes. An outbound AI voice agent can qualify a lead and book the slot inside the same call, then push the appointment into your calendar and CRM without anyone rekeying it. It works best on people who already contacted you: new leads, quote follow-ups and rescheduled jobs. It works poorly as a cold appointment setter dialing strangers, where pickup rates collapse and consent requirements bite hardest.
Get it in writing regardless of what the current minimum is. FCC rules require prior express written consent for artificial-voice telemarketing to mobile numbers, though the Fifth Circuit held in February 2026 that the statute itself requires only prior express consent, oral or written. That ruling binds Texas, Louisiana and Mississippi only, and conflicts with FCC regulation elsewhere. Written, timestamped, number-specific consent keeps you compliant everywhere. This is general information, not legal advice.
Only if they are expecting you. Hiya's 2026 survey of over 12,000 consumers found 86% do not answer unknown numbers, and 20% said no piece of information would make them more likely to answer a business call. The levers that work are a verified business name on caller ID, chosen by 29% as the thing most likely to earn a pickup, calling within minutes of the person contacting you, and never calling people who did not.
Yes, and it is one of the strongest use cases because the recipient already has a booking with you. A conversational reminder can confirm, reschedule on the spot or capture the reason for cancelling, which a one-way text cannot. We looked at what the clinical evidence supports, including where phone reminders beat SMS, in our guide to reducing no-shows.
No. Cold-dialing people who never contacted you is where TCPA and Do Not Call exposure concentrates, pickup rates are lowest, and brand damage is highest. Purchased lists rarely carry consent that survives scrutiny. AIEmply does not support cold outbound campaigns against bought lists. Point the same technology at your own leads, customers and no-shows instead, where consent is documented and people expect to hear from you.
Self-serve platforms bill per connected minute, plus whatever your own build and upkeep time is worth. With AIEmply, outbound is included in the Growth AI Employee at $399/month ($3,348/year), covering full outbound calling, follow-up campaigns, lead re-engagement, Meta ad-form follow-ups, an AI website chatbot and 250 minutes at $0.25 per extra minute. Scale is $599/month with 400 minutes at $0.20. Yearly billing saves 30% and billing starts only after your AI Employee is live.