Remember the old phone menus? "Press 1 for sales, press 2 for support." People hated them, and for good reason. Voice AI is a very different thing. It can listen to someone speaking normally, understand what they want, answer in a natural voice and take action, such as booking an appointment or checking an order.
That's a real change, and a lot of businesses are asking whether it's ready for their phone line. Our answer: for some jobs, yes. For others, not yet.
How it works
A typical voice assistant chains a few pieces together. Speech recognition turns the caller's voice into text. A language model works out what they mean and what to do. It may look up information or call another system, such as your booking calendar. Then a text-to-speech voice reads the reply back. All of this has to happen quickly enough to feel like conversation, because long pauses are what make callers hang up.
Where it works well
Simple, repetitive calls. Opening hours, directions, order status, and "can I change my appointment?" make up a big share of calls to many businesses, and the answers are clear.
Booking and scheduling. Clinics, salons, garages and restaurants lose time and revenue on calls nobody can answer during busy hours. An assistant that books into a real calendar, 24 hours a day, can be a real help.
After hours and overflow. Answering when your team is closed or all lines are busy beats voicemail. At minimum, it can take a message and a call-back number reliably.
Call triage. Working out who's calling and why, then handing over to the right person with a short summary, saves your staff time and the caller's patience.
Where it struggles
Noise, accents and bad lines. Phone audio is poor quality. Strong accents, background noise, mixed languages and people speaking over each other trip up recognition. Test with the accents and languages your callers actually use, because demos are recorded in perfect conditions.
Interruptions and awkward pauses. Humans cut in, change their mind midway and say "um." Assistants vary a lot in how gracefully they cope.
Emotional or complex calls. An upset customer, a complaint, a delicate medical or financial situation needs a person. The assistant must recognise this and hand over quickly.
Confident mistakes. The same problem as with text AI applies, and it's worse when a caller can't see the screen. A wrong appointment time spoken confidently is still wrong.
Names, numbers and spellings. Phone numbers, email addresses and unusual names are error-prone. A good design reads them back for confirmation.
Things to get right
Say it's an AI. Many places have rules or expectations about telling callers, and people mostly don't mind if the service is good. Pretending is a trust risk.
Always offer a way to a human. "Say 'agent' at any time" should work every time. Nothing damages trust like being trapped.
Privacy and consent. Calls may be recorded or transcribed, which can involve legal requirements about consent and data retention that differ by country. Check the rules that apply to you and the provider's data handling, especially for health or payment details.
Don't take card numbers by voice unless the provider has a proper, compliant setup for it. Sending callers to a secure payment link is safer.
Limit its powers. Let it book and check, but require extra confirmation for cancelling, refunding or changing account details.
How to pilot one
Pick one narrow call type, such as appointment booking. Listen to a sample of real calls first to see what people really ask. Run the assistant on after-hours calls only at the start, so the risk is low. Review every transcript for the first couple of weeks. Track a few things: how many calls it completes without help, how many it hands over, and how callers react.
Compare against what you do now. If missed calls fall and your team gets more time, you have a case for expanding. If callers keep asking for a human, fix the script or stop.
Questions to ask a provider
- Which languages and accents have you tested, and how does it perform on phone-quality audio?
- How quickly does it respond?
- What happens when it doesn't understand?
- Where is call data stored, for how long, and who can access it?
- Can it connect to our calendar or system?
- What does it cost per minute, and what happens when usage grows?
Our take
Voice AI is good enough to take over the boring, predictable calls and give your team their time back, as long as you keep a person within reach for everything else. Start narrow, listen to real calls, and expand only when the results back it up.
