A voice-AI startup whose agent answers a clinic's phone like a person — behind a page that promised "human-like" instead of proving it.
Specimen record
Field
Voice AI · healthcare
Engagement
Workshop, then landing page
Timeline
3 weeks, kickoff to live
Baseline → after
−34%bounce on the new page
Where the walk began
Human-like conversational AI for healthcare
"Human-like" is a claim every voice product makes and no visitor believes. Meanwhile the product could do the one thing no claim can: be heard. The old page described the voice at length — and never once played it.
The walk, retraced
The same four lanterns — what happened at each one.
01
Discover
Clinic managers didn't talk about AI at all. They talked about the front desk at 8am — three lines ringing, one receptionist, and patients who book with whoever picks up. The buying moment was the missed call, not the model.
"Every call we miss before nine is an appointment someone else gets."— operations lead, first interview
02
Frame
The founders arrived split between "conversational platform" and "healthcare agent". The signed sentence dropped both: it promised the outcome — every call answered — and dared the visitor to check. Whatever the page claimed, the proof would have to be audible.
The sentence, signed"Every call answered, from day one — and you can listen to it happen."
03
Design
The moment of magic is output, so the fold is a recording: a real booking call, two voices, twenty-two seconds. The type stays quiet around it — this page has one job, and the job is press play.
◦ the shipped heroEvery call answered. Listen for yourself.▸ A real call, 0:22
The hero, as shipped — the recording where the claims used to be04
Ship
Live with analytics on the player itself. Bounce fell by a third — visitors who once left at "human-like" now stayed to the end of a call. The one measured iteration added a second recording: the hard one, a reschedule.