What is IVR?
IVR (interactive voice response) is a phone system that answers callers with recorded or synthetic speech and routes them by keypad or voice. The caller picks an option from a menu, and the system sends the call to a team, a queue or a recorded message. The menu is written in advance, and it recognises only the options it was given.
How does IVR work?
A call opens with a greeting, either a recorded file or text read out by a synthetic voice. Numbered options follow: press 1 for sales, 2 for support, 3 for opening hours.
The caller answers on the keypad or by voice, and the system checks the input against the menu. A match sends the call to a team, a person, a queue, voicemail or another menu. Without a match, the options are read again.
The menu itself is a flow. Each step is a prompt, a wait for input and a set of branches that depend on the answer. Opening hours, out of hours and public holidays can each get their own route. By day the call reaches the team; at night the same number plays a message and may offer voicemail.
- A greeting is played, recorded or synthetic.
- The caller presses a key or says a defined word.
- The call goes to a team, a person, a queue or voicemail.
- Opening hours, out of hours and holidays get separate routes.
What is the difference between keypad and speech input?
Each key on a phone sends a tone signal known as DTMF (dual-tone multi-frequency). ITU-T Recommendation Q.23 sets out the format. A key press plays two frequencies at once, one from a low group of four and one from a high group of four. Q.23 defines sixteen signals: the ten digits, star, hash and four keys marked A to D.
On internet calls, key presses travel as separate events rather than as sound. IETF RFC 4733 explains why: low-rate voice codecs cannot be relied on to reproduce the tones well enough for automatic recognition. Sent as an event, a key press arrives whatever the audio quality.
With speech input, the caller says a word instead of pressing a key. A classic IVR recognises only words and phrases defined in advance, and it does not follow open conversation. The W3C VoiceXML standard calls that list a grammar: the things a caller may say to take an action or give information.
A word outside the grammar is not recognised, and the options are read again. Silence runs out a timeout, and the menu restarts or the call takes a default route. VoiceXML treats the two cases as separate events, each with its own message.
- Keypad: each key sends a two-tone signal.
- Speech: only words in the grammar are recognised.
- An unrecognised word sends the caller back to the options.
- Silence runs out the timeout and the call takes a default route.
Where does IVR get stuck?
The first problem is a question with no option. A menu covers the few most common reasons for calling. A caller asking about price, availability or an order has no obvious key to press. The caller guesses, lands with the wrong team or hangs up.
The second is time. Options are read one by one, so the caller listens until the right one comes up. Sub-menus stretch the wait: a three-level menu means three lists before the right team answers.
The third is the record. A keypad menu captures only the key that was pressed. What a caller wanted before giving up is written down nowhere, so the questions the menu misses stay invisible.
The last is upkeep. A new service or new opening hours means a new prompt and a rearranged flow. When that work waits, the menu reads out information that is no longer true.
- A question with no option lands with the wrong team.
- Sub-menus lengthen the route to the right team.
- The question of a caller who gives up is never recorded.
- Every change means a new prompt and a new flow.
How is an AI voice assistant different from IVR?
An IVR asks the caller to choose; an AI voice assistant listens to the question. The caller explains in their own words, the speech becomes text, and the answer comes from the sources the business supplied.
The two do different jobs. A keypad menu exists to send the call to the right place. An AI voice assistant answers common questions itself: opening hours, the address, the service list and appointment requests. Where its sources have no answer, it makes nothing up and passes the call to a person with a written summary.
Cloud phone platforms already put the two side by side. Some let a menu step hand the caller to a bot that understands speech. Others extend the routing menu with AI agents that guide callers in conversation, without a fixed menu.
- IVR asks for a choice; the assistant listens to the question.
- The menu knows its options; the assistant turns free speech into text.
- The assistant answers only from the sources the business supplied.
- An unanswered question goes to a person with a summary.
When does IVR fit, and when does an AI voice assistant?
IVR works well where calls split into a few clear reasons. In a company with several departments, one key separates sales, support and accounts. Account and order numbers are also taken on the keypad.
An AI voice assistant fits where questions vary and the answers already sit with the business. In a salon, clinic, hotel or field service team, callers tend to ask about hours, the address, a price range or a free slot. Those answers do not fit a menu, but the business already holds them.
The two can also run together. The menu's first step splits calls by team, and one option leads to the assistant. Or the assistant answers every call, and only the topics on its handover list go to a team. ALTAI Digital agrees that list with the client during the build.
- A few clear call reasons across several teams: an IVR menu.
- Numbers keyed in, such as an order reference: keypad input.
- Varied questions with answers already on file: a voice assistant.
- Both together: the menu splits teams, the assistant answers questions.
What this term means for your business
AI voice assistant: the phone rings and nobody misses it
Read the definitionAI chatbot: answers from your sources, hands the rest to your team
Read the definitionChat and voice assistant
Questions on your website, on WhatsApp and on the phone are answered at once with your own information; anything without an answer comes to you.
Sources
- ITU-T Recommendation Q.23: Technical features of push-button telephone sets
- IETF RFC 4733: RTP payload for DTMF digits, telephony tones and signals
- W3C Voice Extensible Markup Language (VoiceXML) 2.0
- Microsoft Learn: overview of Teams voice applications
- Amazon Connect: Get customer input flow block
- Twilio: TwiML Gather documentation
IVR Frequently asked questions
The terms overlap. An auto attendant, as Microsoft Teams calls it, is the menu that routes calls to people and teams. IVR is the wider term and also covers self-service steps, such as keying in an account number.
A classic IVR with speech recognition recognises only the words and phrases defined in advance. It does not follow free conversation, and a word outside its list sends the caller back to the options. Understanding the caller's own sentence is what an AI voice assistant does.
One level can hold as many options as there are digits on the keypad, but the caller waits while each one is read. Options are ordered by the most common reasons for calling and kept few. A route to a person is left for anything the menu does not cover.
That depends on the calls. If they split into a few clear reasons, a menu does the job. If questions vary and the answers sit in your own material, an assistant can finish the call itself. The two can also share one line.
Opening hours, out of hours and holidays each get their own route. Outside hours the system plays an information message and may offer voicemail. The message then waits until somebody listens to it.


We write the proposal after we have found the problem together
In a thirty-minute call we listen to how you work and talk through where time and customers are slipping away. If you do not need a digital fix, we will tell you that too.
30 minutes · no preparation needed