An IVR (interactive voice response) system plays recorded prompts to callers and routes them according to the keys they press on their phone. It announces something like “for bookings press 1, for pricing press 2,” then transfers the caller to the right desk or plays a recorded message based on the number chosen. More recently, AI phone receptionists that take down requests in conversation have come into use as well. This article explains how IVR works, how it differs from an AI phone receptionist, and how to choose between them.
How IVR works
- When a call comes in, the IVR answers automatically and plays a recorded or synthesised prompt
- When the caller presses a key, the IVR reads the tone. These are DTMF tones, made of two frequencies played together, as defined in ITU-T Recommendation Q.23
- Depending on the key, the IVR moves to the next step you configured: transferring to a desk, playing another prompt, recording a message or announcing out-of-hours information
- Most IVR systems log how many calls came in and how often each menu option was chosen
Some IVR systems also recognise short spoken words such as “booking” or “pricing” instead of keypad presses.
Where IVR works well
- A few clearly separate kinds of calls. Bookings, pricing and fault reports, for example, where routing to a handful of desks is all you need
- Many calls that end with an announcement. Opening hours, location or how busy you are can be answered by a fixed message
- Separate desks per department. After routing, a person takes over
On the other hand, callers tend to hang up before reaching the right desk when the menu has many levels. And without a destination for “none of the above,” calls end up back at the main line anyway.
How an AI phone receptionist differs
An AI phone receptionist combines speech recognition, a language model and speech synthesis to talk with the caller and take down their request.
| Aspect | IVR | AI phone receptionist |
|---|---|---|
| What the caller does | Presses number keys | Says what they need |
| Requests it can route | The few options on the menu | Requests in the caller’s own words |
| Taking details | Usually not (voicemail aside) | Name, callback number, request and more |
| Answering questions | Fixed announcements only | Answers registered FAQs in conversation |
| How staff are informed | Call transfer | Summary by email or chat, or transfer |
| Typical errors | Wrong key, lost in the menu | Mishearing (mitigated by reading back) |
| What you prepare | Menu design and recorded prompts | Role, fields to collect and FAQs |
The key difference: an IVR decides where a call goes, while an AI phone receptionist works out what the call is about and records it. For companies that cannot pick up right away, or where a small team handles the phones, transferring a call helps little when nobody is free to answer. Taking the details and passing them on is the better fit.
Ways to combine them
- Route with the IVR, send some options to AI. “Press 1 for bookings” goes to the AI receptionist; everything else is transferred to staff as before
- AI takes details and transfers only urgent calls. The AI listens to the request and transfers to a person only for urgent matters or calls that meet conditions you set
- AI only outside business hours. During the day, the IVR and staff handle calls; after hours, AI takes details for a callback the next business day
What to check when choosing
- What your calls are about. Write down a few days of recent calls and look at the kinds of requests and how many ended with a simple answer
- Your current number and equipment. Whether you keep your existing number or get a new one, and how it connects to your cloud PBX or IVR
- Where the details go. Email, chat or your CRM
- Recordings and records. Retention period, who can see them, and how callers are informed
- Handover conditions. When the AI or IVR passes a call to a person, and whether someone is available at that time
SHANNON products
Our voicast is a voice AI agent platform for developers. AI answers phone, web and app audio, takes down names, phone numbers and requests, and returns the results and a summary by API and webhook. It uses your SaaS’s own phone numbers and lines (a Twilio account, a PBX or a cloud PBX), and after the conversation it can hand the call back for the next step, such as transferring to staff. For businesses in Japan that want AI to answer their own number, AI BizCall answers calls to a dedicated 050 number 24 hours a day, 365 days a year, and sends a summary to the people in charge. Pricing is on each product’s website.
References
- ITU-T Recommendation Q.23: Technical features of push-button telephone sets
- Twilio Docs: TwiML Voice
FAQ
Should we choose an IVR or an AI phone receptionist?
If you have only a few kinds of calls and asking callers to press a number is enough to route them, an IVR will do. If requests vary widely and you want names, callback numbers and details taken down as well, an AI phone receptionist fits better. You can also combine them: route first with the IVR and let AI handle the intake.
How many levels should an IVR menu have?
The deeper the menu, the more callers hang up before reaching the right place. Put the most common requests on the first menu, keep the options to a handful, and always provide somewhere to go for callers whose request does not fit, such as an operator or a callback queue.
Can an AI phone receptionist mishear callers?
Yes. Call quality and speaking style can cause names or numbers to be misheard. Compensate by having the AI read back what it heard, keeping recordings and transcripts, and handing over to a person when it cannot understand.