A new report from AI Voice Research found that enterprise voice agent deployments grew 340% year on year, with 67% of Fortune 500 companies now running production systems.
The global voice agent market reached 47.2 billion dollars in 2025 and is growing at 34% annually.
And the cost? Around 7 to 12 cents per minute. A fraction of what a human call handler costs.
Hello, and welcome to AI SIX. I’m Joanne Sweeney, and I’m on a mission to upskill half a million people in AI, in just six minutes a day.
What is a voice agent?
An AI voice agent is a system that can hold a real-time phone conversation with a human.
Not a recorded menu where you press one for sales.
A real conversation.
It listens. It understands. It responds in a human-sounding voice. All in about a second.
The technology works in three layers:
1. The ears
Speech-to-text converts what the caller says into text.
2. The brain
A large language model, such as GPT, processes the text and decides what to say.
3. The mouth
Text-to-speech converts that response back into spoken words.
This entire loop happens in under a second. If it takes longer than roughly 800 milliseconds, the conversation starts to feel unnatural. Speed is everything.
How are businesses using voice agents?
There are two main use cases: inbound and outbound.
Inbound
On the inbound side, voice agents handle:
- Reception duties
- Customer support
- Frequently asked questions
- Appointment booking
- Call routing
These are calls that already come into your organisation. Instead of voicemail or long hold times, the caller gets an immediate, natural conversation.
Outbound
Outbound is where the economics change.
At eight to twelve cents per call, businesses can now do things that were never financially viable with human teams:
- Follow-up calls after delivery
- Reactivation campaigns for lapsed customers
- Appointment reminders
- Lead qualification at scale
One example shared on the Social Media Examiner podcast described a car wash business using a voice agent to call thousands of churned customers with a reactivation offer. A task that would have taken months was completed in days.
What do you need to get started?
The good news is you do not need to be a developer.
Platforms such as Retell AI, Vapi, and ElevenLabs now offer no-code builders. You select your speech-to-text provider, your language model, and your voice, and the platform connects them. You can build a working demo in an afternoon.
But the technology is not the most important part.
Planning is.
Before building anything, map your use case:
- What types of calls do you receive?
- What are the most common questions?
- What actions should the agent take? Book an appointment? Log a support ticket? Transfer to a human?
Think in conversation paths, not scripts. The agent must handle the unexpected, not just follow a decision tree.
Build in confirmation steps. The agent should repeat back key information before taking action.
And design for failure. There must always be a clear escalation path to a human.
Now, governance
If you operate in the EU or serve EU customers, transparency rules under the EU AI Act come into effect in August 2026.
Under Article 50, if your AI system interacts with a human, you must disclose that they are speaking with AI. Your voice agent must announce itself at the start of every call.
GDPR applies to any personal data processed during the call. Voice data can be classified as biometric data if used for speaker identification.
The ePrivacy Directive requires informed consent for call recording and automated marketing calls.
In the United States, the FCC confirmed in February 2024 that AI-generated voices fall under the Telephone Consumer Protection Act. All existing robocall consent requirements apply.
Penalties range from 500 to 1,500 dollars per call, with no cap. A 10,000-call campaign with compliance failures could result in 15 million dollars in exposure.
The legal landscape is fragmented and evolving. The Colorado AI Act takes effect in 2026 with additional requirements for high-risk AI systems. Several US states now require or encourage disclosure when customers interact with AI.
If you are deploying voice agents, legal advice specific to your jurisdiction is essential.
Practical steps
If you are considering voice agents, start here.
1. Audit your phone operations
What calls come in? What calls go out? Where are the bottlenecks?
2. Start with inbound
Customer support and appointment booking are typically the lowest-risk, highest-return use cases. Perfect this before moving to outbound.
3. Plan compliance from day one
Build disclosure into every call. Document consent. Store call logs securely and redundantly.
4. Test relentlessly
Call the agent yourself. Try to break it. Refine the prompts, conversation paths, and escalation triggers.
5. Update your AI acceptable use policy
Define what the voice agent can and cannot do. Set boundaries around data access. Review the policy quarterly.
The bottom line
The technology is ready. The cost is negligible.
But governance must keep pace.
If you are deploying AI agents that speak directly to your customers, you must be as deliberate about compliance, disclosure, and escalation as you are about voice quality and speed.
Voice AI is no longer experimental.
It is answering the phone.
📚 AI Playbook on Amazon, the perfect podcast companion.
📰 Human in the Loop, my LinkedIn Newsletter for weekly AI insights.
🎓 AI Playbook Diploma, CPD Accredited, structured learning to understand, use, and govern AI.
📞 Book a call at digitaltraining.ie
Key sources:
Social Media Examiner, “AI Voice Agents: How to Get Started” by Tommy Chryst and Michael Stelzner (20 January 2026)
AI Voice Research, “The State of Voice Agents in 2026” (December 2025), enterprise deployment data
Market.us, Voice AI Agents Market report, $2.4B (2024) to $47.5B (2034) at 34.8% CAGR
FCC Declaratory Ruling, February 2024, AI-generated voices classified as “artificial” under TCPA
EU AI Act, Article 50 transparency obligations, effective August 2026 (digital-strategy.ec.europa.eu)
Telnyx, “EU AI Act: Compliance Essentials for Voice AI in Europe” (September 2025)
Henson Legal, “AI Voice TCPA Compliance: Legal Guide for Voice AI” (January 2026)
No-code platforms referenced: Retell AI (retellai.com), Vapi (vapi.ai), ElevenLabs (elevenlabs.io)
Text-to-speech alternatives: Cartesia (cartesia.ai)
Speech-to-text recommendation: Deepgram (deepgram.com)
About the Podcast Host
AI and Communications Specialist | Author | Podcaster | International Speaker
Joanne Sweeney is a recognised leader in digital marketing and digital transformation, specialising in the practical application of AI to enhance communication strategies. The founder of Digital Training Institute, Joanne, empowers leaders and communications professionals to leverage AI responsibly and effectively, ensuring transparency, trust, and tangible results.
She completed the Harvard Kennedy School Executive Programme in AI and Public Policy this summer before launching her fourth book AI Playbook, in September. Joanne has also published two editions of Public Sector Marketing Pro, which were preceded by her first title, Social Media Under Investigation Law Enforcement and the Social Web.
Her books have become go-to resources for governments, public sector and policing agencies, as well as marketing and communications professionals, all navigating the challenges of the changing Internet.
Joanne is the host of the AI SIX Podcast, a daily six-minute podcast that provides actionable insights, expert interviews, and practical advice to help professionals navigate the new world of work.
