Home>Blogs>PCI-Compliant Phone Payments with AI Voice Agents

PCI-Compliant Phone Payments with AI Voice Agents

PCI-Compliant Phone Payments with AI Voice Agents

PCI-Compliant Phone Payments with AI Voice Agents

The safest way to take card payments in an AI voice agent call is to keep card data away from the AI entirely.

Hand the caller to a secure payment capture step, usually keypad (DTMF) entry with a PCI-validated provider, and return only a token and result to the agent.

This guide explains the design. It is not a compliance assessment; confirm scope with your Qualified Security Assessor (QSA) and acquirer.

Why the AI Should Never Hear the Card Number

If a caller speaks a card number, it can pass through speech recognition, the language model, transcripts, logs and recordings. Each of those systems may then fall into PCI DSS scope.

Keeping card data out of the AI path is a design choice that can reduce scope. The PCI Security Standards Council document library holds the current standard and guidance.

Reference Design

StageWhat happensCard data present?
ConversationAgent confirms identity and amountNo
HandoffAgent starts secure captureNo
CaptureCaller enters digits on keypad; provider captures themYes, within provider
Tokenize or chargeProvider sends to gatewayYes, within provider
ReturnAgent receives result, token and last 4 digitsNo
Wrap-upAgent confirms payment outcomeNo

Step 1: Use a Secure Capture Service

Programmable voice platforms offer capture services. For example, Twilio Pay collects card details by DTMF and sends them to a payment connector. Twilio documents that it can tokenize or charge, and returns limited details such as the last four digits.

Check each provider's current PCI attestation and what it covers.

Step 2: Mask DTMF Tones

Keypad tones can be decoded from recordings. Use capture that suppresses or masks tones so agents, recordings and the AI never receive them.

Step 3: Pause or Exclude Recording and Transcription

Stop recording, transcription and AI audio streaming during capture, or confirm your provider handles this automatically. Test it, then check sample recordings.

Step 4: Handle Spoken Card Numbers

Some callers will read their number aloud. Plan for it:

• Prompt clearly: "Please use your keypad. Do not say your card number."

• Detect and redact: apply number-pattern redaction to transcripts and logs

• Do not store raw audio of the payment segment where avoidable

• Escalate callers who cannot use a keypad to an approved alternative

Our guide on PII redaction for LLM applications covers detection patterns.

Step 5: Return Only Safe Data to the Agent

The agent should receive payment status, a token or reference, amount and last four digits only. Never pass the full number, expiry or security code to the model or CRM.

Step 6: Log and Monitor

Log handoff start, capture result, retries and failures without card data. Alert on failed captures and unusual retry patterns.

Worked Example: Utility Bill Payment Call

Illustrative example. A utility's voice agent handles bill payment calls.

• Conversation: agent verifies the account and reads the balance

• Handoff: "I'll connect you to secure payment. Please enter your card number on your keypad."

• Capture: the telephony provider's secure capture tokenizes the card with the gateway

• Return: agent receives "success, ending 4242, reference ABC123"

• Wrap-up: agent confirms and offers a text receipt

Recordings and transcripts show a gap during capture, verified in weekly sample checks.

What to Confirm with Your QSA

• Which systems remain in scope with this design

• Whether your capture provider's attestation covers your use

• Recording, transcription and log handling during capture

• Requirements for staff who can access call data

• Compensating controls for spoken card numbers

Common Mistakes to Avoid

• Letting the AI collect digits: brings STT, LLM and logs into scope

• Forgetting transcripts: recording is paused but transcription is not

• Storing full payloads in CRM notes: keep only token and last four

• No fallback: callers without keypads need another approved route

Limitations

PCI DSS scope depends on your full environment. This guide does not certify compliance. LuMay does not claim PCI DSS certification; payment capture relies on your chosen validated provider.

Next Step

If you want a voice agent that takes payments without card data reaching the AI, the Voxentis voice agent team can design the handoff with your telephony and payment providers.

• Twilio Pay TwiML documentation

Tags:

#AI Voice Agent#Voice Automation#Inbound Calls#Contact Centers

Frequently Asked Questions

Everything you need to know about Voice Agent

1. Can an AI voice agent take card payments?

Yes, by handing off to a secure capture service so the AI never receives card data.

2. Is DTMF capture required?

It is a common approach. Your QSA can advise on alternatives for your environment.

3. Does using a PCI provider make my whole system compliant?

No. It can reduce scope, but your remaining systems and processes still need assessment.

4. What should the agent receive after payment?

Status, token or reference, amount and last four digits only.

5. What if a caller speaks their card number?

Redact transcripts and logs, avoid storing that audio, and remind callers to use the keypad.

6. Should calls be recorded during payment?

Pause or exclude recording and transcription during capture, and verify it works.

7. Which telephony platforms offer secure capture?

Several do, for example Twilio Pay. Check each provider's current attestation.

8. Can the AI read back the card number?

No. Read back only the last four digits.

9. Who decides PCI scope?

Your QSA and acquirer, based on your full environment.

10. Does LuMay hold PCI DSS certification?

No such claim is made. Payment capture uses your validated provider.

11. How do I test the design?

Run test payments, then review recordings, transcripts and logs for any card data.

About the Editorial Team

Sarath Babu

Sarath Babu

Content Writer and SEO Specialist at Lumay

Creates insightful content on SEO, AI-powered marketing, digital growth, and emerging technologies. He simplifies complex topics into practical, research-backed guidance.

Palanisamy

Palanisamy

CEO and Founder at LuMay

27+ years of experience leading enterprise-scale AI, data, and systems architecture initiatives, delivering mission-critical platforms with a strong emphasis on trust, governance, and reliability.