Skip to main content
June 24, 2026 · By Piyush Sahoo · Updated November 3, 2026 A multi-level IVR is an automated phone menu where a caller’s choice can open another menu instead of ending at one destination. Each layer narrows the reason for calling, so one business number can reach many teams, languages, and self-service answers without a receptionist. This guide helps you decide whether you need one, and how deep it should go. It covers what the research says about menu depth, how to build a menu from your own call log, and what India’s newest telecom rule can teach any business.
Key takeaways
  • Layers have a cost. Every level is more listening and one more chance of a wrong turn, so each one has to earn its place.
  • Two levels is a sensible default. In our illustrative model, a third level saved listening time only past about 40 destinations.
  • “Five options at most” is weaker advice than it sounds. A 2008 study found that a broad menu beat a deep one on both performance and satisfaction.
  • Your call log should shape the menu. Grouping reasons by volume and answering the top one automatically can help more than adding branches.
  • India now has a reference design. TRAI’s November 2, 2026 amendment sets out language choice, request type, sub-menu, a human option, and a queue call-back for telecom providers.

What is a multi-level IVR?

IVR stands for interactive voice response, the system behind “press 1 for sales”. A single-level IVR asks one question and acts on the answer. A multi-level IVR can ask a follow-up, the way a good receptionist would. You may also hear it called a nested IVR, a multi-tier IVR, or a phone tree. They all describe the same idea. The diagram shows one for an imaginary online store. Tree diagram of an illustrative phone menu for an Indian online store. The caller chooses Hindi, English, or Tamil, then track my order, returns and refunds, or buy or ask. Only returns has a third level. Paths end in a status read out or a connection to a team, and pressing 0 reaches a person on every level. Tree diagram of an illustrative phone menu for an Indian online store. The caller chooses Hindi, English, or Tamil, then track my order, returns and refunds, or buy or ask. Only returns has a third level. Paths end in a status read out or a connection to a team, and pressing 0 reaches a person on every level. A public Indian example shows the idea at national scale. Business Today’s 2021 report on the Rail Madad 139 helpline listed these details at launch:
  • Languages. The helpline was available in 12 languages.
  • Urgent help first. Pressing 1 for security or medical help went straight to a call centre executive.
  • A sub-menu for enquiries. Pressing 2 opened further choices such as PNR status and train times.
  • A way out. The star key reached an executive directly.
That description is more than five years old, so the live menu may differ. The design choice still holds: an emergency never waits behind a tree.

How does a multi-level menu work on a call?

Every level repeats the same three steps:
  1. The system speaks. It plays a recording or reads text aloud.
  2. The caller answers. They press a key or, if the menu allows it, say a word or two.
  3. The system decides. It either plays another menu or finishes the path.
A path can finish in one of three ways. The system can read out an answer, such as an order status. It can connect the caller to a team, using rules like those in our guide to call routing. It can also take a message for later. Keypad menus don’t need an app or a smartphone, because a push-button phone sends a standard tone for each key. The ITU’s recommendation on push-button telephone sets covers these technical features. That reach matters in India, where your callers may be on any kind of phone.

When do you need more than one level?

You need another level when callers must make a choice before the main question makes sense. These are the usual reasons:
  • Language. The caller can’t use a menu they don’t understand.
  • Location. A chain may need to route by city or region before it routes by team.
  • Customer type. A new customer and an existing one may need different teams.
  • Volume of destinations. A keypad has twelve keys, and you need some of them for “repeat” and “talk to a person”.
A growing organisation chart isn’t a reason on its own. Callers don’t know how your teams are divided, and they shouldn’t have to learn. Decision map. If there is one language and up to about nine destinations, use a single menu. Otherwise, if callers would not recognise their reason in a short menu label, use speech or an AI agent with keys as a fallback. If they would, and one branch still holds too many destinations after two levels, add a third level on that branch only. Otherwise use two levels. Decision map. If there is one language and up to about nine destinations, use a single menu. Otherwise, if callers would not recognise their reason in a short menu label, use speech or an AI agent with keys as a fallback. If they would, and one branch still holds too many destinations after two levels, add a third level on that branch only. Otherwise use two levels. This table compares the three shapes:

How many levels and options should a menu have?

The usual advice is two or three levels and about five options. That advice is a reasonable place to start, but it deserves a closer look.

What the research says

The five-option limit is often justified by a famous 1956 memory paper. A 2008 study in the journal Human Factors argued that the paper doesn’t support the limit. The researchers gave people functionally equivalent IVRs, one broad and one deep, and asked them to complete email tasks. The broad design won on both performance and satisfaction. The gap was largest for people with lower working memory capacity. The lead author’s dissertation reached the same conclusion: shorter menus stacked more deeply were harder to use. This doesn’t prove that longer menus are always better. It is one experiment, on a speech-enabled system, and its tasks were about email rather than customer support. Treat it as a caution, not a rule for your callers. It does mean you shouldn’t add a level just to shorten a list.

A simple way to compare shapes

We built a small model to compare listening time. It assumes each option takes 3 seconds to say, each move to a new menu takes 2 seconds, and every destination is equally likely. Grouped bar chart from an illustrative model of average listening seconds. Four destinations: single menu 7.5, two levels 11. Nine destinations: single menu 15, two levels 14. Sixteen: two levels 17, three levels 20.5. Thirty-six: two levels 23, three levels 23.5. Sixty-four: two levels 29, three levels 26.5. Grouped bar chart from an illustrative model of average listening seconds. Four destinations: single menu 7.5, two levels 11. Nine destinations: single menu 15, two levels 14. Sixteen: two levels 17, three levels 20.5. Thirty-six: two levels 23, three levels 23.5. Sixty-four: two levels 29, three levels 26.5. Three things stand out:
  • At four destinations, a sub-menu wastes time. Two levels take 11 seconds on average, against 7.5 for one menu.
  • At nine, it’s nearly a tie. One menu averages 15 seconds and two levels 14. The worst case differs more: 27 seconds against 20.
  • A third level pays off late. It trails two levels at 16 destinations and roughly ties at 36. It moves ahead only past about 40, and it’s clearly ahead at 64.
The model flatters deep menus, because it ignores wrong turns. A caller who picks the wrong branch has to back out and listen again. The chance of that rises with every extra decision. So start with two levels, and use a third only on the one branch that needs it. If a menu seems to need a third level everywhere, try speech input or fewer destinations first.

How do you design a menu from your call log?

GetVoIP’s advice to map a month of inbound calls by reason is the right first step. The next step is to group those reasons in words a caller would use. Illustrative call log. Where is my order 36 percent and change delivery 8 percent become key 1, track my order, at 44 percent. Refund or return 20 percent and payment problem 14 percent become key 2, returns and payments, at 34 percent. Product question 12 percent and new order 6 percent become key 3, buy or ask, at 18 percent. Anything else, 4 percent, goes to key 0, talk to a person. Illustrative call log. Where is my order 36 percent and change delivery 8 percent become key 1, track my order, at 44 percent. Refund or return 20 percent and payment problem 14 percent become key 2, returns and payments, at 34 percent. Product question 12 percent and new order 6 percent become key 3, buy or ask, at 18 percent. Anything else, 4 percent, goes to key 0, talk to a person. Then work through these steps:
  1. Answer the top reason automatically. If the system recognises the caller’s number and can read out an order status, many callers finish there.
  2. Put the biggest group first. Using the example shares above and 3 seconds per option, popularity order averages about 5.5 seconds of listening. The reverse order averages about 9.5. At 1,000 calls a day, that illustrative gap comes to about 68 minutes of caller time daily.
  3. Say the reason before the key. “For refunds, press 2” lets a caller ignore the options that don’t apply to them.
  4. Keep special keys consistent. A common choice is star for “go back” and 0 for an operator on every level. Consistency means callers learn your menu only once.
  5. Handle silence and wrong keys kindly. Repeat the menu a couple of times, then pass the caller to a person. Don’t hang up on them.
  6. Offer a call-back when queues are long. It replaces hold time with a promise you can keep.
  7. Move offers out of the way. TRAI found that plans and promotions at the start of a call cost consumers time. A customer calling with a problem shouldn’t have to sit through your latest offer.
  8. Test every path. Include wrong keys and after-hours calls, as GetVoIP also advises.

What should you measure afterwards?

Averages hide where callers get stuck, so look at each branch separately:

What do India’s rules and callers change?

India adds two pressures to menu design: more languages, and a new regulatory benchmark.

TRAI’s new reference design

On November 2, 2026, TRAI notified the Telecom Consumers Complaint Redressal (Fourth Amendment) Regulations, 2026. It sets out what automated systems on a telecom provider’s Consumer Care Number must offer. The IVR and human-support clauses take effect 90 days after publication in the Official Gazette, and the language clause takes effect after 30. This rule binds telecom service providers, not your business. It is still the clearest public statement of what India’s telecom regulator expects from a good menu. This table is our reading of what you could borrow, and it isn’t legal advice. The amendment’s analysis is worth reading too. TRAI said that, in many instances, consumers had to go through several layers of menus before reaching the right category. Some stakeholders asked for a human option at every level, but TRAI settled on the third stage as the balance between automation and support. A 0 key on every level goes further than that, which is why we suggest it. This builds on an earlier structure. TRAI’s 2024 consumer explainer already showed three steps: language, category, then sub-category or an agent.

Language is the first level for many Indian businesses

Scroll’s analysis of the 2011 census put Hindi speakers at nearly 44% of India, a figure that includes languages such as Bhojpuri. That data is old, but the point stands: Hindi and English alone leave many callers out. TRAI’s amendment asks telecom complaint centres to offer at least one other official language of the states in the service area, where applicable, in addition to Hindi and English. Banks are moving the same way. In December 2025, the government told the Rajya Sabha that banks run multilingual contact centres, and that RBI wants customer communications in Hindi, English, and the regional language. That report concerns language, not menu depth, so ask your compliance team what applies to you. A language menu takes up one of your levels, so plan for these points:
  • Speak each option in its own language. A Tamil speaker should hear the Tamil option in Tamil.
  • Offer languages your team can support. A Tamil menu that ends with a Hindi-only agent breaks the promise it just made. Read-out answers work without a matching agent, so they are a good place to begin.
  • Remember the choice. A returning caller recognised by their number can skip the language step.
  • Use a separate number where it helps. A regional number can start in that region’s language and save everyone a level.
  • Budget for upkeep. Every prompt change now means a new recording in every language.

If the menu plays on calls you place

This guide covers people calling you. A “press 1” menu on outbound calls raises different questions, because India’s commercial calling rules apply. TRAI amended those rules again on September 18, 2026. Start with India calling regulations on Vobiz and how to manage spam complaints, and get consent in place first.

Should you use a multi-level menu or an AI voice agent?

A menu makes the caller do the sorting. An AI voice agent lets them say “I want to check my refund” and works out the rest, which can replace several levels in one sentence. Most businesses don’t have to pick only one. A sensible mix looks like this:
  • Keep keys for exact choices. Language selection, “press 1 to confirm”, and a route to a person all suit the keypad.
  • Use speech for varied requests. It helps when reasons don’t fit a short list, or when callers don’t know your terms.
  • Keep a human route either way. TRAI’s amendment allows AI agents on telecom complaint lines, but with human oversight and a clear notification to the consumer.
Vobiz is the telephony layer in this mix, not the agent. It carries the call and the audio, while the AI comes from you or a partner such as Vapi or Retell. Our page on AI voice agents explains how these integrations fit together.

How Vobiz handles multi-level IVR

Vobiz gives you the building blocks for a menu, and your software decides what each caller hears. You’ll need a developer or a partner to set this up. They can start with how VobizXML works, the call-control language behind every Vobiz menu.
  • Every press reaches your software. Vobiz sends each choice to your application through webhooks (callbacks), and your application replies with the next step. Your tree can change by caller, time of day, or agent availability. The cloud IVR overview describes the approach.
  • Prompts can be recordings or text-to-speech. Play streams an MP3 or WAV file, so you can record prompts in Hindi, Tamil, or any language your callers use. Speak reads text aloud in its listed languages.
  • Keypad and speech can share one menu step. Gather can listen for both, and it acts on whichever arrives first. It also takes hints for expected phrases.
  • Keypad input carries no speech charge. Speech recognition is billed by the duration analysed, while key presses aren’t charged for detection. See Vobiz pricing for current rates.
  • Caller recognition uses the caller ID. Each request to your application includes the From number, so you can look up a returning caller and skip steps.
  • Sensitive entries can stay out of logs. Gather has a log setting that turns off logging of digits or speech for that step, which suits account details.
  • Agents are one transfer away. Call transfer and call escalation move the caller to a person at any level.
  • Your data shows where callers drop. Your software sees each press, and call records and analytics data fill in the rest.
To see the parts working, look at the IVR menu example or the visual XML builder. Before you go live on an Indian number, complete India number KYC. Test calls use your own credentials and can be billed, so confirm the numbers you dial. If you’d like help scoping a menu, talk to the Vobiz team.

Frequently asked questions

It is an automated phone menu where a choice can open another menu instead of ending at one destination. Each layer narrows the reason for calling, so one number can reach many teams, languages, and self-service answers.
A single-level IVR plays one menu and acts on the choice straight away. A multi-level IVR can follow that choice with another menu. Use one level when a few options cover everything. Add a level when callers must first pick a language, a location, or a broad group.
Start with two levels. Add a third only on the branch that needs it. In our illustrative model, a third level saved listening time only past about 40 destinations. Your own call log should decide the final shape.
A multi-level IVR is one menu with layers. Multiple IVRs are separate menus, usually chosen by the number dialled or the time of day. You can combine them, for example with a daytime menu and a shorter after-hours menu, each with its own levels.
Yes, if the system supports speech. On Vobiz, one menu step can listen for a key press and speech together, and it acts on whichever comes first. Keep the keypad available, because it works on any phone and in noisy places.
For telecom providers’ Consumer Care Numbers, yes. TRAI’s November 2, 2026 amendment requires a route to a human representative within each request category and a queue call-back option. It does not bind other businesses, and your own sector may have separate rules, so check with your compliance adviser.

Further reading on Vobiz

Sources

Build on Vobiz

Provision a number and build a multi-level IVR with nested menus in minutes