Add tap-to-confirm to a Retell AI voice agent
powsoo_show_card with the transcribed value, the caller taps Confirm or edits a character, and Powsoo fires a webhook with the final value. Setup is one web component plus one script tag and does not change how Retell runs the conversation.Where a Retell agent loses data
Retell is an English-first, US-focused voice-agent platform with compliance built in. It handles the speech-to-text, model, and text-to-speech pipeline for you, which is convenient. But the capture problem is inherent to speech recognition, not to any one platform. Rare tokens like names, emails, and numbers are exactly what word-error-rate-optimised models protect least.
Phone numbers are the sharpest example. State-of-the-art over-the-telephone digit recognition reaches about 98% per digit, but accuracy compounds across a string: 0.98 to the tenth power is roughly 0.82, so about one ten-digit number in five contains at least one wrong digit. The agent reads it back confidently and the caller has to catch the error by ear.
Add on-screen confirmation with Powsoo
- Connect your Retell account in the Powsoo dashboard by pasting your Retell API key, which Powsoo encrypts server-side.
- In your Retell agent, add a server-side custom function named
powsoo_show_cardpointing at Powsoo’s endpoint, withspeak_during_executionandspeak_after_executionboth off so the agent pauses silently while the card is on screen. - Append Powsoo’s system-prompt snippet so the agent calls the function instead of asking callers to spell values back.
- Create a widget in Powsoo, then drop the
<powsoo-voice widget-id="wid_…">web component and the Powsoo widget script onto the page hosting the call. - When Retell calls the function, Powsoo renders the card, the caller taps Confirm or edits a character, and the confirmed value flows back to the agent and on to your backend.
Which fields to confirm
- Phone numbers: the highest failure rate on digit strings.
- Email addresses: the same on-screen pattern applies to emails.
- Dates: resolve ambiguous spoken dates to a concrete, tappable value.
Related
Frequently asked questions
Does Powsoo work with Retell’s custom functions?
Yes. You add powsoo_show_card as a server-side custom function on the agent, pointing at Powsoo's endpoint. When Retell decides a field needs confirming, it calls that function with the transcribed value, and Powsoo renders the card in the browser.
Why do phone numbers fail on Retell voice calls?
Digits are short and acoustically confusable (O vs zero, fifteen vs fifty), and telephone bandwidth makes it worse. Whole-number accuracy compounds: at 98% per digit, a ten-digit number is only about 82% correct end to end.
What about phone-only Retell calls?
A PSTN call has no browser to render into. Use an SMS fallback so the caller confirms on their phone. See the tap-to-confirm vs SMS fallback guide.