Designing the voice AI that answers when you call, and calls when you don't
Client
hiRobin.ai
Year
2026 - Present
Founding designer for hiRobin, an AI that actually talks on the phone. I own onboarding, monetization design, and growth end to end, took signup-to-conversion from ~30% to 80%+.
Scope of Work
Impact
30–40% → 80%+
Signup-to-activation conversion
74%
Peak D7 retention after redesign
End-to-end
Data dive through shipped redesign
Hero image placeholder
The problem
Early signup numbers were bad, and everyone had a theory. The engineering team believed the system — including OTP delivery — was working fine. Their failure tracking only counted one kind of OTP failure: someone typing the wrong code. It did not track whether the OTP was ever sent at all, or whether it arrived late. I didn’t think it was that simple, so I went and looked myself.
Funnel drop-off placeholder
The clearest issue was not only a low completion rate, but a measurement system that could not describe why people were leaving. Before changing the experience, I needed a more truthful view of the funnel.
The diagnosis
I pulled funnel data, session recordings, and support tickets, then mapped actual behavior against what the team’s failure logs claimed was happening. The gap was immediate: users were dropping at the OTP step in ways the existing tracking could not even see. I found three distinct failure modes — OTP never sent, OTP delayed, and OTP entered wrong — while only the final case was being measured.
There was also a smaller but real friction point: no auto-copy on the OTP. Users had to leave the app, open Messages, memorize or copy the code, then come back — an extra round trip that cost people mid-flow.
Failure modes placeholder
The pushback
Presenting an untracked failure mode to a team confident in its own system is not a data problem, it is a trust problem. My first pass got a polite but skeptical response. I kept returning with sharper evidence — not “OTPs are failing” as a vague claim, but the exact breakdown of never-sent, delayed, and wrong-entry cases. Eventually the team agreed to swap the OTP provider from Firebase and see what happened.
Evidence review placeholder
That test broke the deadlock. Once the switch happened, OTP failures dropped sharply and the team could see directly what the tracking gap had been hiding. From there, we moved into a full onboarding redesign together.
Before / after onboarding comparison placeholder
The result
Signup-to-activation conversion moved from roughly 30–40% to 80%+ after the provider switch and the onboarding redesign. D7 retention peaked at 74% in the same window of work.
Where it stands now
I’m currently leading monetization design ahead of hiRobin’s Series A — pricing, packaging, and paywall placement — working directly with the founders on what to gate and how. We’re also preparing usability testing to pressure-test open questions. The team is still actively experimenting, which is exactly what this stage calls for.
What I’d take forward
A team’s confidence in its own system is not evidence. Untracked failure modes look like “everything’s fine” until someone goes looking. Winning a technical disagreement is rarely about being right in one meeting — it is about returning with sharper evidence until “let’s just try it” becomes the easiest option.
Role — Founding Product Designer
Skills — Data-driven diagnosis, cross-functional influence, onboarding UX, growth design