Missed calls are lost customers. A tradesperson up a ladder, a dentist with a full waiting room or a restaurant mid-service can't always answer. AI voice agents, systems that answer the phone, understand natural speech and respond in a natural voice, have become good enough in 2026 to take many of those calls properly.

They're not magic, and badly set up they're infuriating. Here's how to get it right.

What an AI voice agent can do

TaskWorks well?
Answer common questions (hours, prices, location, services)Yes
Book, move or cancel appointments in your calendarYes, with calendar integration
Take detailed messages and send them to you by text or emailYes
Qualify enquiries (job type, postcode, urgency, budget)Yes
Transfer urgent calls to a personYes
Take orders for collectionOften, with a clear menu
Handle complaints or emotional callsPoorly; hand to a person
Give advice with legal, medical or financial consequencesNo

How they work

  1. A call comes in to your number (or is forwarded when you don't answer).
  2. Speech recognition turns the caller's words into text.
  3. An AI model works out what they want and decides what to do, using your instructions and knowledge base.
  4. It takes actions through integrations: checking your calendar, booking, sending a text.
  5. A voice engine speaks the reply.

Modern systems do this fast enough to feel conversational, handle interruptions and cope with accents far better than the old "press 1 for sales" menus.

Where voice agents go wrong

  • No escape route. Callers trapped with a bot that won't transfer them get angry fast.
  • Pretending to be human. It erodes trust when discovered, and in the EU it can break the AI Act's transparency rules.
  • Making things up. Without clear limits, an agent may invent prices or promise availability. Read AI hallucinations.
  • Thin knowledge. If it doesn't know your services, prices and policies, it can't help.
  • Bad audio. Slow responses or robotic voices make callers hang up.

How to set one up that customers like

1. Decide its job

Pick a narrow scope to start: "Answer questions about services and hours, book appointments, take messages for everything else." You can widen it later.

2. Write its knowledge base

Give it:

  • Services with plain descriptions and prices or price ranges.
  • Hours, holidays, address, parking, service areas.
  • Booking rules (lead times, cancellation policy, deposits).
  • Your top 20 questions with your real answers.
  • Things it must never say or promise.

3. Script the opening

"Hi, you've reached Bright Smile Dental. I'm the practice's AI assistant. I can book or move appointments and answer questions, or take a message for the team. How can I help?"

Short, honest, useful.

4. Set handover rules

Transfer or escalate when:

  • The caller asks for a person.
  • It's an emergency or complaint.
  • The agent can't answer after two attempts.
  • The topic involves money disputes, health or legal matters.

Out of hours, take a detailed message and promise a callback time you can keep.

5. Connect your tools

Calendar for bookings, CRM for caller history, text messaging for confirmations. Without integrations, an agent can only talk.

6. Test with real calls

Call it yourself in different ways: mumbling, interrupting, asking odd questions, changing your mind. Ask friends with different accents. Fix the knowledge base based on what goes wrong.

7. Review transcripts weekly

Read call summaries and transcripts. Look for wrong answers, missed handovers and questions it couldn't answer. Each one is a quick improvement.

Privacy and compliance

  • Tell callers they're speaking to AI and that calls may be recorded.
  • Check where call data is stored and processed, and sign a data processing agreement with the provider. Read using AI tools without breaking GDPR.
  • Keep recordings only as long as needed.
  • Don't take card details through the AI unless the provider is properly certified for payments. Send a secure payment link instead.

Is it worth it?

A rough way to decide:

QuestionYour number
Calls missed per week
Share that were potential customers
Share of those you'd likely win
Average value of a new customer
Monthly cost of the voice agent

If even two or three extra customers a month cover the cost, it usually pays. Our AI automation ROI guide explains how to estimate this properly.

Voice, chat and email together

Many customers prefer to message rather than call. A website chat agent that answers from the same knowledge base keeps answers consistent across channels. Zelvorm Assist answers customers from your website and documents and hands over to your team when it should. For a broader view, see AI chatbots for small business: what they can and can't do.

The short version

AI voice agents can answer calls 24/7, book appointments, take messages and qualify enquiries. Start with a narrow job, give it a solid knowledge base, say upfront that it's AI, set clear handover rules, connect your calendar and review transcripts weekly. Never let it take payments or handle complaints alone.

Want one set up for your business? Our AI automation team builds voice and chat agents connected to your systems.