AI voice agents in customer service: how they work and why they already handle 65% of support requests

Back to blog

Two years ago, 52% of inbound customer support requests were resolved without a human. This year that figure is 65%. [GetNextPhone 2025] In a report published in March 2025, Gartner predicts that by 2029, AI will independently handle 80% of routine customer service situations. The forecast specifically concerns recurring, structural service requests, not all contacts. [Gartner 2025]

What is a voice agent?

A voice agent is an AI system that holds a phone conversation with a person. It listens to speech, understands the content, chooses a response, and speaks it back in a natural voice, in real time.

This isn't a traditional IVR system ("press one for billing, press two for technical support"). A voice agent understands free-form language and responds based on context.

A modern voice agent is built on three technologies:

• Speech recognition (speech-to-text): converts speech into text in real time
• Language model (LLM): understands context and produces a response that fits the situation
• Speech synthesis (text-to-speech): converts text into natural-sounding speech

All of this happens in under a second. To the caller, the experience feels like a normal phone conversation.

What are they doing at real companies?

These cases differ in type, and their results shouldn't be generalized directly against each other.

A voice agent replacing a human agent: Everise, an international customer service firm, deployed an AI voice agent in partnership with Retell AI. [Retell AI 2024] The results were measurable:

• 65% of calls are handled entirely by AI
• Wait time dropped from 5–6 minutes to zero
• Savings: 600 person-hours

Moving from an IVR menu to a voice agent: Doxy.me, a telehealth service, replaced its traditional button-based menu with a voice agent. The share of calls handled by IVR was 5%. The voice agent now handles 30%. This figure specifically reflects the shift from an outdated system, not a starting point of zero.

AI in broader customer service: Klarna resolved customer service requests with its AI agent 82% faster than a human agent. Resolution time dropped from 11 minutes to 2 minutes, and customer satisfaction stayed the same. Klarna's case mainly concerns text-based customer service, not voice agents specifically. It illustrates AI's broader potential in customer service.

Market growth shows the direction

The voice agent market was worth $2.4 billion in 2024. Market.us projects it will grow to $47.5 billion by 2034. [Market.us 2024] This is a market research firm's forecast, not a realized market size, but the direction is clear.

Market forecasts are worth reading as directional. What matters more is that businesses see measurable results in their own operations, and that's what drives investment.

What do customers actually think?

This is where the tension lies. According to a SoundHound AI study, 84.7% of people say they prefer a human agent before trying the alternative. [SoundHound 2024] But when the same study exposed respondents to a real voice agent experience, satisfaction more than doubled. 44% then said they preferred AI for most service situations.

Experience changes the opinion. Expectation doesn't match reality.

Voice agents work especially well in situations where the customer wants a fast answer or needs to handle a routine matter without waiting on hold. From the SoundHound study:

• 50.4% find AI helpful when it routes them to the right person
• 49.6% prefer AI for delivery and order confirmations
• 46.9% prefer AI for managing appointments

When doesn't a voice agent fit?

Honesty builds trust: voice agents don't fit everywhere. Situations where a human is the better option:

• Complex complaints requiring judgment, negotiation, or exception decisions
• Sensitive personal data: health, financial, or legal matters where the customer needs to feel trust
• Legal or contractual decisions that require an accountable person
• Urgent emergencies that require immediate human action
• Emotional crisis situations where the customer needs empathy, not a process

A good voice agent recognizes these situations and routes the caller onward rather than trying to handle everything itself.

An example from a Finnish SME

A large share of property maintenance calls concern fault reports, urgency assessment, and gathering contact details. A voice agent fits these situations well: it captures the details, assesses urgency based on the customer's description, and routes urgent cases onward.

The same logic works for an HVAC business, a cleaning service, or a dental clinic: routine requests (appointments, quote requests, fault reports) get handled automatically. Demanding situations go to a human.

How do I get started?

Getting started doesn't require a big project. A practical starting model has three steps:

1. Pick one call type. An appointment, a fault report, or a quote request. One clear task at a time.
2. Decide your escalation rules. In which situations does the call go to a human? Complaints, sensitive matters, and emergencies always get routed onward.
3. Measure the result. Calls answered, leads captured, appointments booked, and time saved. The first week already shows the direction.

Puhebotti as a voice agent

Puhebotti is a Finnish voice agent built for the needs of SMEs. It answers calls 24/7, holds a natural conversation, and delivers a clear summary to the business owner after every call.

It doesn't replace human service in complex situations. It makes sure every caller gets an answer when the business owner or staff can't answer themselves.