Keypad input: when callers should press a key instead of speaking

Voice agents are good at conversation. They're less reliable with a 12-digit order number read out over a noisy line, or a PIN the caller would rather not say out loud in a shop. For those moments, the phone's keypad is still the best input there is.

Turn it on

On the agent's Call tab, switch on keypad input. Keys the caller presses then reach the agent as if they had said them, in a form the agent understands:

[keypad: 4 2 0 1 7 #]

Keys pressed close together arrive as one message, and # sends what was typed at once. You can set how long Vaakyo waits between keys before it sends them (2.5 seconds by default). Pressing a key while the agent is talking interrupts it, so a caller who already knows the drill doesn't have to wait.

The agent's prompt is told the format, so you only have to ask for it in plain words.

When to ask for keys

Use the keypad for things that are long, exact or private:

• Order, booking and account numbers. "Please type your 6-digit booking number, then press hash."
• PINs and OTPs. Callers often don't want to say them aloud.
• Quick choices. "Press 1 to confirm, 2 to cancel." This is handy for callers who prefer it, but don't build the whole call as a menu. Let them answer in words too.
• Ratings. "On a scale of 1 to 5, press the number that fits."

Keep speech for everything else. Names, addresses and "what can I help you with?" are conversations, not key presses.

Write the prompt for both

Callers will mix the two. Some will say "four two zero one seven", some will type it, and some will type half of it and say the rest. Tell the agent to accept either:

Ask for the order number. The caller may say it or type it on the keypad.
Read it back digit by digit and ask them to confirm before you look it up.
If it isn't 8 digits, ask again once, then offer to help without it.

Reading back and confirming matters more with numbers than anywhere else. One wrong digit sends a tool call to the wrong order.

See what was pressed

Each keypad entry is a call.dtmf event on the call's timeline, with the digits pressed, and webhook endpoints can subscribe to it. Together with analytics fields (for example order_id as text, or rating as a number), what the caller typed ends up as structured data on the call.

Test it

The browser test call in the console has its own keypad, so you can try the whole flow before a real call. Then place a phone call and try it the way your customers will: one-handed, in a hurry, pressing hash too early.