👉 Voice AI -> Capturing Secure Keypad Input (DTMF) from Caller 📱
G
G J
Payments taken during phone calls is my top most priority but I've also been thinking of other use cases such as asking callers to authenticate themselves using their keypad by entering OTP sent by text message to the phone number on file, multi-step verification (e.g., enter postcode, or DOB digits + last 4 of phone number), or a preset secret code on their profile - for secure requests. There are many possibilities..
- Multi-level IVR-like flows, but powered by AI context (e.g., “Press 1 to reschedule, 2 to cancel, 3 to speak to an agent”)
- Numeric field data capture: Enter postcode, date of birth, account number, item quantity or amount (e.g., “Enter the number of units you’d like to order”)
- Order-related enquiries (press to select order type (e.g., “1 for new order, 2 for existing order”)
- Important changes to the contact's information, such as date of birth, etc.
- Enter ticket number or reference ID
- Mortgage/loan: Enter loan number or last 4 digits of account number
- Insurance: Enter policy number or claim reference
- Healthcare: Enter patient ID or appointment reference
- Enter card number, expiry, and CVV via keypad (with masking for PCI-friendly secure DTMFinput).
- Confirm payment amount by pressing a key (e.g., “Press 1 to approve £36.00”).
- Select stored payment method (e.g., “Press 1 for Visa ending 42, Press 2 for MasterCard ending 19”)
- Press a key to confirm terms, disclosures, or agreements
Must support retry logic
: “That code didn’t match. Please try again.” Fallback
: After 3 failed attempts, transfer to a member of staff. Sequential capture
: "Enter your 6-digit policy number, followed by the pound key." Inline validation
: "That number doesn’t match our records. Please re-enter." Masking & Security
: - Ensure sensitive inputs (card numbers, PINs) are not stored in transcripts.
User Experience
: - Allow “#” or “*” as end-of-input markers for variable-length entries.
- Provide clear instructions: “Enter your 6-digit code, then press #.”
- Short confirmation prompts: “You entered 1234. Press 1 to confirm, 2 to re-enter.”
Fallback & Escalation
: - If keypad input fails repeatedly, offer to send a text message with a link or live agent transfer.
- Timeout handling: “I didn’t receive any input. Would you like me to repeat the options?”
Log In
S
Sales & Marketing
Merged in a post:
AI Agent DTMF Capability Needs to be a thing- Before it launches!
K
Kurt Sweet
Hi! Native DTMF support should be a core capability of our AI agent This is what bridges the gap between modern automation and the legacy infrastructure that still runs the business world.
Relying on middleware or third-party workarounds to send tones introduces latency, complexity, and failure points. All of which undermine the simplicity and reliability clients expect from our AI and costs us more as Agencies to employ- I've kept my AI's 100% native, until now because without this I'm forced to use an integrator.
In practical terms, DTMF isn’t a “nice-to-have”; it’s the difference between an agent that talks and one that acts. It enables fully autonomous call handling... transfers, confirmations, and menu navigation without breaking the flow or the illusion of competence.
If the AI agent is to be trusted in professional environments, especially where multiple departments or legacy phone systems are in play, native DTMF is non-negotiable.
Even so much as a .wave file that plays the tones based on user input would be sufficient...
Thanks
M
Market Suite.io
⚠️ DEV TEAM - Quick Win Implementation Request
## PROBLEM: Voice AI currently forces speech-only input. Users need the option to enter numeric data via keypad (DTMF) instead of/alongside voice - especially for long number sequences where STT accuracy degrades.
## SOLUTION: Enable Twilio's existing
<Gather input="speech dtmf">
for Voice AI workflows. This runs both input channels in parallel - whichever completes first wins. DTMF path bypasses STT entirely and passes clean digits directly to the AI context.## WHY THIS MATTERS:
- Data integrity: 16-digit credit card + exp + CVV + zip = 28 digits. Voice recognition fails, user repeats, AI confirms, still wrong, loop continues. DTMF = zero ambiguity.
- Call efficiency: Numeric entry via keypad is 3-5x faster than speaking digits + readback confirmation loop
- User choice: Older demographics/enterprise users expect keypad option. Current either/or split (Voice AI vs IVR) forces abandonment of one workflow paradigm.
## HIGH-VALUE USE CASES:
- Credit card collection (PCI compliance + accuracy)
- Phone number capture (10-11 digits)
- OTP/2FA code entry (6 digits)
- SSN/FEIN/Account numbers (sensitive PII)
- Birthdate (8 digits in MMDDYYYY)
## IMPLEMENTATION:
Since LC Phone is Twilio underneath, this is merely config exposure, not new dev. Just need UI control in Voice AI settings to toggle DTMF collection on/off during prompts. Twilio handles all the heavy lifting.
## BOTTOM LINE: Table stakes feature for enterprise voice AI. Eliminates false choice between conversational AI and data entry accuracy.
Thanks,
JT @MarketSuite.io