How an AI-Powered Onboarding Cut Churn by 18% at a HealthTech Startup

AI · 6 min read

The startup’s onboarding previously used a one-size-fits-all form: lengthy, clinical, and low in engagement. Designers partnered with ML engineers to prototype an LLM-driven conversational onboarding that asked fewer but more context-rich questions, inferred priorities, and suggested a starter care pathway. The design challenge was to keep conversations short and transparent while avoiding clinical misadvice.

Product decisions included limiting the LLM’s role to clarification and triage, surfacing sources for any medical-sounding suggestion, and always routing high-risk responses to a clinician review queue. The UX used progressive disclosure: initial text prompts, followed by quick-choice buttons, then a summary confirmation screen that captured consent and next steps.

In A/B testing over three months, the AI-onboarding cohort completed onboarding 36% faster and had an 18% lower 30-day churn compared with controls. Qualitative feedback highlighted perceived empathy and relevance as the main gains. The team’s takeaway was that LLMs can add value in early-stage onboarding if designers treat them as augmentation layers with strict safety boundaries and explainability baked into the UI.