Yes. A business can automate phone calls with AI today, and Benian Technologies builds voice agents that answer inbound calls and work outbound lists, but the two directions live under very different rules. Inbound calls the customer chose to make mostly raise disclosure and recording consent questions. Outbound automated phone calls with an AI voice fall under the federal Telephone Consumer Protection Act (TCPA), and in February 2024 the FCC ruled that AI generated voices count as artificial voices under that law.
So the real question is not whether the technology works. It is which calls you have the right to make, to whom, at what hours, and with what records behind them. Get that wrong and one campaign can create more legal exposure than the leads were worth. Get it right and an agent can call every lead you already own within minutes of the lead arriving, which a small staffed team rarely manages during busy hours.
This page covers what to automate first, the US rules that apply to AI outbound calls, how to work a lead list you already hold, and how to watch and stop a campaign. It is general information, not legal advice. Have counsel review your consent language and calling plan before the first dial.
Inbound versus outbound: which calls to automate first
Start inbound. Calls that arrive after hours, at lunch or while staff are with a customer are the cheapest win: the caller already wants you, consent questions are simpler, and every call the agent answers is one that used to reach voicemail. The good first jobs are booking and rescheduling, answering the same ten questions about hours, location and process, collecting intake details, and transferring anything unusual to a person with a summary.
Outbound comes second, and the safest outbound calls go to people who asked to hear from you: a web form lead who requested a callback, a customer confirming an appointment, a past client whose quote went quiet. Cold AI telemarketing to purchased lists is the opposite end. It carries the most legal risk and can damage how carriers label your phone numbers, and we do not recommend starting there.
A simple test for any call type: if a human on your team would need a script, a lookup in one system and a yes or no outcome, an agent can usually handle it. If the call needs judgment about money, a complaint or a medical or legal question, keep a person on it and let the agent route the call instead.
The US rules for AI outbound calls
The TCPA restricts calls that use an artificial or prerecorded voice. The FCC's February 2024 declaratory ruling made clear that a voice generated by AI, including a cloned or synthetic voice, is an artificial voice for this purpose. In practice, an AI voice agent placing outbound calls is treated like a robocall, however natural it sounds.
What that means for consent depends on the call. A marketing or sales call to a mobile or residential number with an artificial voice generally needs prior express written consent: a signed agreement, which can be electronic, that clearly authorizes calls from your business using an artificial voice to that number. An informational call, such as an appointment reminder or a reply to a request the person made, generally needs prior express consent, which is a lower bar but still has to exist and be provable.
Several states have their own telemarketing laws, sometimes called mini TCPAs, with stricter consent, hour or frequency rules. Rules also change. Treat the consent record as the asset: for every number, store when consent was given, the exact wording the person agreed to, the page or form it came from, and the number it covers.
Consent, Do Not Call lists and calling hours
Before any sales call, scrub the list against the National Do Not Call Registry and against your own internal do not call list. The internal list matters as much as the federal one: when anyone tells your agent or your staff to stop calling, that request has to reach the list and stop every future dial from every number you use. Federal rules restrict telephone solicitations to between 8 a.m. and 9 p.m. in the called person's local time, and some states narrow that further or limit weekend and holiday calls.
Engineering handles most of this. The agent should read the lead's time zone from the area code or address, refuse to dial outside the allowed window, check the do not call status at the moment of dialing rather than at list upload, and cap attempts per lead. An existing business relationship can matter for some calls, but it does not replace written consent for artificial voice marketing calls, so do not lean on it.
Disclosure and recording consent
Artificial voice calls must identify the business making the call at the start and give a way to reach it. For sales calls, the person also needs an easy way to opt out during the call. We build agents to say plainly that they are an AI assistant calling on behalf of the business, and to treat any version of stop, remove me or do not call as an opt-out, read by the model rather than matched against a word list.
Recording is a separate rule set. Some states require every party on the call to consent to a recording. If you record or transcribe calls, and you should for quality review, the agent states that the call is recorded near the start. Both disclosures cost a few seconds and remove a category of risk.
Working an existing lead list: an outbound agent's first week
The best outbound list is usually the one already in your CRM: new inquiries no one called back fast enough and dormant leads that went quiet months ago. E-Ihracat Turkiye did exactly this. The first version of the voice agent Benian built pulled every new and dormant lead from the team's CRM and called them, and the client reports 15 meetings booked in the first week, 10 from new leads and 5 from retargeted ones, with 4 of those meetings closing. E-Ihracat operates in Turkiye, so the US rules on this page describe what a US business would need, not that engagement.
A first week usually runs like this. Day one: export the list, remove numbers without a usable consent record, scrub against do not call lists and tag each lead new or dormant. Day two: test calls to staff phones, reviewing every transcript. Days three to five: a small batch of real leads, with a person reading each transcript the same day and fixing the script. Then widen the batch only if the opt-out rate and transfer quality look right.
Callbacks are where automated phone service often breaks. If the lead says call me Thursday after 3, the agent should write that time to the CRM as a task or schedule a new call, and the attempt counter should know the person asked for it. Meetings go straight onto a calendar the team already uses, with a summary of what the lead said.
Monitoring and stopping a campaign
Watch five numbers daily: connect rate, opt-out requests, transfers or meetings booked, calls flagged by staff as wrong, and any complaint. A rising opt-out rate usually means the list or the message is wrong. A falling connect rate can mean carriers are starting to label your numbers as spam, which also hurts the calls your staff place by hand.
Every campaign needs a stop switch that one person can use without an engineer, and a rule for when to use it: a complaint that mentions consent, a spike in opt-outs, or any call where the agent said something untrue. When should you not automate outbound calls at all? If you cannot show a consent record for most of your list, if the list was bought, or if you would only call a few dozen people a month, a person with a good script is the better choice. Cost on our side is scoped to the work and driven by call volume, the number of systems the agent must read and write, languages, and how much review the first weeks need.