“About 3 minutes”
Patients wanted to know what they were signing up for before they started. So the intro leads with a time estimate, before the first question, not buried later.
2024 · AI health-tech · Solo designer
Redesigning an AI medical assistant to keep patients talking
Previsit.ai runs the intake chat that happens before a doctor’s appointment. A patient answers questions about symptoms, medications, and history in a chatbot, so the visit starts with the basics already covered.
The chat only works if people finish it. And nearly half of them didn’t. They’d start, quit partway through, and leave the intake half-done.
57% → 77%Conversation completion rate
3 → 4Answer quality, rated by doctors (out of 5)
77Questions answered by patients who finished
My role
Sole product designer at Previsit.ai, responsible for all design decisions across the product. For this project: conversation analysis, patient interviews, flow redesign, UI, copy, and interaction design.
- User Experience
- User Interface
- Research
- Conversation design
- Copywriting
Before & after
before: chat state · 57% finished
Hi! Please answer the following questions:
The old intake was one message. A generic hello, then every question dumped at once, no sense of how long it would take or how far you’d got. Patients opened it, saw a wall, and left.
after: chat state · 77% finished
An intro that names your doctor. Time estimate. One question at a time. Progress bar.
The insight
I went in assuming the chat asked too much. Reading 30 conversations and interviewing 6 patients told me otherwise.
People stayed when they could see the end coming. Around question 6 was where they started leaving, and it held whether the questions were easy or hard. The finish line mattered more than the length.
Constraint
That set the constraint: 6 questions from the doctor, plus up to 2 AI follow-ups when an answer was too vague to use. Which broke the obvious way to show progress. You can’t count “3 of 8” when the total keeps changing.
the cliff at question 6
Patient completion rate
4 / 30 reach the end
Three fixes that only work together
The time estimate sets expectations up front. The progress bar keeps people moving through a question count that isn’t fixed. The follow-ups happen inside that bar, so a clarifying question never feels like extra work piled on.
The bar never moves backward
A fixed count like “3 of 8” breaks the second the AI adds a follow-up. Now it’s “3 of 9” and the finish line just moved, so trust goes with it. A bar keeps moving forward no matter how many questions get asked.
Follow-ups that stay inside the bar
“I smoke” is useless without the frequency. So the AI asks one targeted clarification when it’s medically needed, and the bar keeps moving the whole time. The patient never sees the goalposts shift.
In a medical context a vague answer is often worse than none. “I take medication” is useless without the dosage. So the AI catches the gap and asks one follow-up, only when it actually matters. The patient just sees the bar move, not a counter ticking up.
follow-up demo
57% → 77%Conversation completion rate
Measured across 47 patient conversations in the two weeks after launch. The people who used to quit at the wall were now reaching the end.
Reflection
A time estimate, a progress bar that only moves forward, and follow-ups that stay inside it took completion from 57% to 77%. People finished because they could always see how close they were.
This one was mostly about restraint: taking things off the screen until the patient felt sure about what was left. If you’re building patient-facing AI and fighting drop-off, I like these problems. Say hi.
natalia.wlwsk@gmail.com →