AI Glossary Customer service

Interactive Voice Response (IVR)

IVRnoun · pronounced "eye-vee-ar"

  • Tidio Agents team
  • Updated November 19, 2026
  • 6 min read
  • 7 sources
Definition

Interactive voice response (IVR) is a telephone system that answers calls automatically and lets callers choose options by pressing keys or speaking, then plays information, completes a simple self-service task, or routes the call to the right team. It is the "press 1 for billing" layer in front of most contact centers.

IVR was built to sort calls cheaply before a person picks up. It does that well for a few predictable choices. It does badly when the caller's problem does not fit the menu, and every caller pays for that in time.

Voice AI changes the front door. Instead of a menu, a voice agent asks "How can I help?", understands the answer, and either resolves the request or transfers the call with the reason already known. Most companies do not rip out the IVR to get there: the agent usually sits behind it or replaces its first menu.

How it works

DTMF and speech IVR

The original IVR listens for keypad tones. Each key on a phone sends a pair of audio frequencies, known as dual-tone multi-frequency (DTMF) signaling, standardized by the ITU for push-button telephones.1 On internet telephony the same key presses travel as events inside the audio stream.3 The IVR maps each tone to a menu branch.

Speech IVR adds a recognizer for short spoken answers: "billing", "yes", an account number. The W3C's VoiceXML standard describes such dialogs, combining synthesized speech, recorded audio, and recognition of both spoken and DTMF input, and was written to bring web-style development to IVR applications.2 Speech IVR still follows a tree. It only accepts the words each node expects.

CallGreetingMenuSub-menuQueueno match: repeat menuagent picks upDTMF tones · short spoken keywords
A typical menu IVR. Every failed choice sends the caller back a level, and the reason for the call is often lost before a person answers.
Problem

Why callers dislike IVR menus

The complaints are structural, not a matter of voice quality or menu wording.

  • Options arrive one at a time. A caller must hold five spoken choices in memory and guess which one fits, without seeing them.
  • The menu mirrors the company, not the problem. "Billing", "technical", and "sales" are departments. A caller with a wrong charge after a plan change belongs to two of them.
  • Wrong choices are expensive. A misroute means a transfer, a new queue, and often a second identity check.
  • Context is lost. Whatever the caller entered may not reach the person who answers, so they repeat the account number and the story.
  • The route to a person is hidden. Callers learn to press zero or say "agent" repeatedly, which defeats the menu's purpose.

Research on customer effort points the same way. The authors of the Customer Effort Score suggest using it alongside operational measures such as repeat calls, transfers, and channel switching.7 A menu that misroutes callers drives up all three.

Example

One call, menu IVR versus voice agent

Menu IVR

Routes
Caller(Presses keys) My flight tomorrow was canceled and I need to get on another one.
  1. Menu"For new bookings press 1, for existing bookings press 2, for baggage press 3." Caller presses 2.
  2. Sub-menu"To change a booking press 1, to cancel press 2." Caller presses 1.
  3. InputEnters the booking reference on the keypad, letters by key.
  4. QueueWaits for the rebooking team. The agent asks for the reference again.

Routed, then handled by a person. The caller did the sorting work, and the agent repeats the identification.

Voice AI agent

Resolves
CallerMy flight tomorrow was canceled and I need to get on another one.
  1. IdentifyMatches the caller's number to a booking and confirms the passenger name.
  2. Toolbookings.alternatives(ref)→ 2 options, same day, no fare difference
  3. OfferReads both options and asks which one the caller wants.
  4. Actbookings.rebook(ref, option_1)→ confirmed, new ticket issued
  5. CloseSends the itinerary by text message. If the caller asks for a person, it transfers with the booking and reason attached.

Rebooked in one call. People handle the cases the policy does not cover.

Illustrative example for an airline or tour operator. Tool names and options are placeholders.

Voice AI

How voice AI agents replace or sit behind IVR

There are three common patterns, and they can be adopted in order.

1

Natural-language front door

IVR keeps: authentication, queues

The first menu is replaced by an open question. Speech recognition and an intent model route the call to the right queue with the reason attached.

2

Agent behind the IVR

IVR keeps: menus for other topics

One branch, such as "existing bookings" or "billing", hands the call to a voice agent that can resolve it. Contact-center platforms support this directly: in Amazon Connect, a bot is added to a contact flow that is then assigned to a phone number.5

3

Agent as the front lineExample above

IVR keeps: fallback and compliance prompts

The voice agent answers every call, resolves what it can, and transfers the rest. On SIP telephony, a transfer uses the REFER method, and the agent passes a summary to the receiving team.4

Either way, the voice agent needs the same things as a chat agent: tools into booking, billing, or order systems; clear rules for when to transfer; and a disclosure that the caller is speaking with an AI system, which the EU AI Act requires unless it is obvious.6

Compared

DTMF IVR, speech IVR, and voice agents

SystemCaller inputUnderstandsCan actBest for
DTMF IVRKeypad tonesOnly the keys offeredScripted lookupsA few fixed choices, PIN entry
Speech IVRShort spoken wordsWords each menu node expectsScripted lookupsHands-free menus
Natural-language routingOpen speechThe reason for the callNo, it routesCutting misroutes
Voice AI agentOpen speech, interruptionsFull requests, several issues at onceYes, through toolsResolving rebookings, billing, claims
In practice

Where voice agents take over from IVR

Telecoms

Billing and plan questions

Explains a bill line by line, applies a correction within policy, and changes the plan, instead of routing between billing and sales.

identify → read bill → correct → confirm
Insurance

First notice of loss

Takes the claim by voice, asks the policy-specific questions, creates the claim record, and sends a reference number.

verify policy → collect → open claim
Travel

Disruption peaks

When flights are canceled, call volume jumps. The agent rebooks within fare rules while people handle the exceptions.

find booking → options → rebook
Retail

Delivery and returns calls

Gives the live delivery status and books a return pickup, for customers who still prefer the phone.

order → track → book pickup

Illustrative examples, not client deployments.

Risks

Risks in replacing IVR, and the controls

Speech errors on account details

ControlConfirm numbers and names back to the caller, and allow keypad entry as a fallback for long identifiers.

Wrong action on the phone

ControlRead the change back before acting and require explicit confirmation for anything that costs money or cancels a service.

No way to reach a person

ControlKeep a spoken or keypad route to a human at every step, and transfer with context instead of to a generic queue.4

Disclosure and recording rules

ControlTell callers they are speaking with an AI system and follow the same call-recording notices the IVR already plays.6

FAQ

Common questions

What does IVR stand for?

Interactive voice response. It is the automated phone system that greets callers, offers options by keypad or speech, and routes or serves them.

What is the difference between IVR and a voice agent?

An IVR follows a menu and only understands the keys or words each step expects. A voice agent understands open speech, asks follow-up questions, and can complete tasks through connected systems before transferring to a person.

Can a voice AI agent work with our existing IVR?

Yes. Most deployments start by connecting the agent to one IVR branch or replacing the first menu, while keeping the existing telephony, queues, and authentication steps.

Is DTMF still used?

Yes. Keypad tones remain useful for PINs, card numbers, and quick yes or no choices, and voice agents often accept them alongside speech.

Sources

References

  1. ITU-T. Recommendation Q.23: Technical features of push-button telephone sets. International Telecommunication Union, 1988.
  2. W3C. Voice Extensible Markup Language (VoiceXML) Version 2.0. W3C Recommendation, 2004.
  3. Schulzrinne, H. and Taylor, T. RFC 4733: RTP Payload for DTMF Digits, Telephony Tones, and Telephony Signals. IETF, 2006.
  4. Sparks, R. RFC 3515: The Session Initiation Protocol (SIP) Refer Method. IETF, 2003.
  5. Amazon Web Services. Add an Amazon Lex bot to Amazon Connect. Amazon Connect Administrator Guide.
  6. European Union. Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50.
  7. Dixon, M., Freeman, K. and Toman, N. Stop Trying to Delight Your Customers. Harvard Business Review, July 2010.

Tidio Agents

Build an agent that resolves, not only answers

We design and build custom AI agents for customer service, shopping and back-office work, on your infrastructure, with the code handed over to you. Projects start with a 21-day private AI audit.

Request a call