# What is IVR (interactive voice response) and how does it work? > What IVR or interactive voice response is, how it works inside (DTMF tones and menus), the types there are, what it does badly, how it differs from an AI voice agent and how a field service menu should be set up. Joaquín Tenés · 2026-11-14 · https://delegic.com/en/glossary/ivr “If you’re calling about a breakdown, press 1. For quotes, press 2.” Almost everyone has heard an IVR, even without knowing it’s called that, and it’s the first thing a customer hears when they call many field service companies. Here’s what it is, how it works inside, the types there are, what it does badly, how it differs from a voice agent and how to set one up. ## In short - IVR stands for interactive voice response. - It answers the call with recorded messages or a synthetic voice and a menu, and routes it based on what’s pressed or said. - It understands keys because each one sounds as two tones at once, an international standard called DTMF. - It routes calls, but it doesn’t resolve the request: at the end of every option there’s a person or a voicemail box. - An AI voice agent goes further: it asks for what’s missing and leaves the request logged. ## What is IVR? [IBM](https://www.ibm.com/think/topics/interactive-voice-response) defines it as an automated telephone system technology that lets callers receive information, provide it or make requests using their voice or a menu. In practice, it’s what answers before anyone else: it greets the caller, offers some options and, depending on what they choose, passes the call to a department, a queue, a voicemail box or a recorded message with information, like the office hours. It isn’t the same as a phone system, though they usually go together. The phone system holds the lines and extensions and knows how to transfer a call from one phone to another; the IVR is the menu that decides where it goes. That’s why it’s usually bought with the [cloud phone system](/en/glossary#cloud-phone-system), as one more feature, alongside schedules and call forwarding. ## How does it work inside? According to IBM, an IVR relies on pre-recorded messages or text-to-speech, and on a tone interface known by its initials, DTMF (dual-tone multi-frequency). Behind it you need a network, some databases and a server that runs the menu application, written in VoiceXML, a language designed for that. The tones are the part almost nobody knows about. When the customer presses a key, the phone doesn’t send a number: it sends a sound made of two tones at once, one low and one high, and the IVR recognizes the key by the pair. Recommendation Q.23 of the ITU, the United Nations agency for telecommunications, sets those frequencies for the whole world: ITU-T Recommendation Q.23, points 5 and 6, on the [ITU](https://www.itu.int/rec/T-REC-Q.23) website, accessed on November 14, 2026. Four low frequencies, one per row of the keypad, and four high ones, one per column. The 5, for example, sounds at 770 and 1336 hertz at the same time. The two tones of each key, in hertz 1209133614771633697123A770456B852789C941\*0#D The standard allows for sixteen keys, the ten digits and six reserve signals. The star and the pound key are on every phone, but the fourth column, A to D, isn’t on an ordinary phone’s keypad. Because the tones travel down the same line as the voice, the IVR hears them just as it would hear a word, which is why they work from any phone, landline or mobile, without installing anything. Those keys are used to build the menu, which is a tree: each option leads to another message, another menu further down or a final destination. In a field service company it can be as simple as the one below, with three branches, and what matters is where each one ends. An IVR can also look things up, not just route. If it’s connected to the company’s database, it can ask for a contract or request number on the keypad and read out its status. When it can’t find what’s asked, as IBM explains, the menu options are used to pass the call to the right person. ## What types of IVR are there? IBM distinguishes three, depending on how the customer chooses. The first is **touch-tone**: the customer presses a key for each option. It’s cheap, always works and doesn’t mishear keys, but it makes you listen to the whole list of options before choosing. The second is **directed dialog**: the system asks and the customer answers with a word from a list, for example “say repairs, quotes or billing”. It saves the keypad, but it’s still a menu, and it fails when the customer says something that isn’t on the list. The third is **natural language**: the customer explains in their own words why they’re calling and the system, using speech recognition, understands the intent and picks the branch. It’s what Google Cloud calls a [conversational IVR](https://docs.cloud.google.com/dialogflow/es/docs/integrations/phone-gateway), and it’s the border with the [AI voice agent](/en/blog/ai-voice-agent): it understands like an agent, but what it does with what it understands is still route the call. ## What does it do badly? IBM lists three drawbacks: menus that are too long or complicated frustrate callers, long waits don’t go away just because there’s a menu in front of them, and the impersonal treatment can make a customer who was already annoyed even more so. All three are familiar to anyone who has called a large company. In field service there’s a fourth that weighs more: the IVR doesn’t resolve the request. It can send the breakdown call to the right queue, but someone has to pick it up, ask for the address, the equipment and the symptom, and type it in. If nobody is there at that hour, the repairs branch ends in the same voicemail box as all the others. And a menu doesn’t know what’s urgent either. The customer chooses “repairs” both for a dripping boiler and for a shop with no heating in January, and the difference only comes out when someone listens to the message or calls back. ## IVR or AI voice agent? The difference is in what’s done when the call ends. An IVR, even a conversational one, decides where the call goes. An AI voice agent does the part that comes after: it asks for what’s missing for that kind of fault, checks whether the customer has a contract and leaves the request logged in the company’s system, with the right urgency. The model behind it is an [LLM](/en/glossary/llm), the same kind of model as in a chat, with speech recognition on the way in and a synthetic voice on the way out. They aren’t incompatible. You can keep the menu for what works well as a menu, like billing or office hours, and put the agent on the repairs branch or outside office hours, when today everything ends in voicemail. How a call with an agent works, step by step, and what it costs, is in [what is an AI voice agent](/en/blog/ai-voice-agent). ## What should a field service menu look like? First, keep it short. Every extra option is time the customer spends listening before they can choose, and a customer with a breakdown has no patience for five options. The repairs branch goes first, because it’s the one that matters most and is most in a hurry. Second, decide what happens outside office hours. Voicemail means someone has to listen to the messages in the morning and call back; forwarding to the [on-call](/en/glossary#on-call) technician gives the emergency to someone who can fix it, but also hands them every call that isn’t one. And third, always leave an option to talk to a person: it’s what the customer looks for when the menu doesn’t have what they need. With [Delegic](/en/channels/voice), the company forwards its number to the agent’s, always or only outside office hours, or sets it as the destination of the repairs branch in the menu it already has. The agent talks to the customer, asks what the technician needs to know and leaves the request in the office’s inbox; if the customer asks for a person, it tells them the office will call back, and the request arrives with their number and what they said. ## Frequently asked questions ### What does IVR stand for? Interactive voice response: the system that answers a call with an automated menu. ### What is DTMF? Dual-tone multi-frequency, the way a phone sends key presses: each one is a pair of tones, one low and one high, that the IVR recognizes. The frequencies are set by ITU Recommendation Q.23. ### Is an IVR the same as a phone system? No. The phone system manages the lines and transfers calls between extensions; the IVR is the automated menu that decides where each call goes. Most cloud phone systems include one. ### Can an IVR understand speech? Some can. A directed-dialog IVR recognizes a word from a list, and a natural-language one understands a whole sentence. But understanding isn’t the same as resolving: it still only routes the call. ## Related terms - [AI voice agent](/en/blog/ai-voice-agent): the one that picks up the call and leaves the request done. - [Cloud phone system](/en/glossary#cloud-phone-system): the one that holds the lines and usually includes the IVR. - [On-call](/en/glossary#on-call): who handles emergencies outside office hours. - [Latency](/en/glossary#latency): how long a voice agent takes to answer. - [Speech recognition](/en/glossary#speech-recognition): what turns the caller’s words into text.