Convarza AI
Call flows

Build an IVR menu

The classic press-1 tree, when it helps. Often an AI agent that asks is better — here is how to tell which you have.

Start by asking whether you need one at all. IVR menus were invented because routing a call required the caller to do the routing. That constraint is gone: an AI agent can ask what they need and act on the answer, in their own words, without anybody memorising four options.

An IVR is still right when the options are few, genuinely stable, and genuinely different destinations — sales versus support versus accounts, in a business where those are separate teams. It is wrong when you are using it to interrogate the caller before helping them.

IVR menu is one of the steps in the picker — and often not the one you want.
IVR menu is one of the steps in the picker — and often not the one you want.

Before you start

  • A call flow, and an honest answer to whether a menu beats an agent simply asking.
  • Your three or four options, ordered by what callers pick most.
  • A path for no input and for a wrong press.

Steps

Decide honestly whether a menu beats an agent asking

If most callers pick the same option, remove the menu and go straight there. If the options need explaining, they are not options — they are a question.

Add an IVR menu step

Each option leads to its own chain of steps, so an IVR is really several small flows sharing an entry point.

That is why menus grow unmanageable: four options with three steps each is twelve things to maintain, and every one of them needs a terminating step.

Keep it to three or four options

People cannot hold more than four in their head. By option five they press zero, and whatever zero does is now your actual routing strategy.

Say the option before the number

"For sales, press one" beats "press one for sales". Callers are listening for their own word, not for a digit — and if the digit comes first they have to hold it in memory while they find out what it means.

Handle no input and wrong input

Somebody will say nothing, and somebody will press seven. Both need a path, usually to a person or to an AI agent. An unhandled wrong press is a caller in limbo.

Put the most common option first

The most common for them, not the most important to you. Ordering by internal priority is how callers end up listening to the whole menu every time.

Ring your own number and walk every branch

Including the ones you are sure nobody uses. Menus that read well often sound interminable, and you only find out by listening.

Check it worked

Time the whole menu from answer to first option being available. If it is more than about fifteen seconds, cut it. Then check every branch terminates in a person, a bot, a message or a hang-up.

Worth knowing

The parts people get wrong

  • Every IVR option needs a terminating step. An option that leads nowhere leaves a caller in silence with no way out.
  • If most callers press the same option, that is data. Consider removing the menu entirely.
Questions

The ones people actually ask

How many options should a menu have?

Three or four. People cannot hold more than four in their head, and by option five they press zero — at which point whatever zero does is your actual routing strategy, whether you designed it or not.

What if somebody presses nothing, or presses seven?

Both need a path, usually to a person or an AI agent. Somebody will do each of these, and an unhandled wrong press leaves a caller in limbo.

How should I word the options?

Say the option before the number. For sales, press one beats press one for sales, because callers are listening for their own word rather than a digit — and if the digit comes first they have to hold it in memory while they find out what it means.

Rather be shown than read?

Thirty minutes, your numbers and your channels on the screen.

No setup fee. No per-conversation charge. Cancel from inside the product.