Conversational Onboarding: How Tutorly Used LLMs to Personalize the First 5 Minutes

AI · 6 min read

Tutorly's original onboarding asked users to fill a long preference form before browsing tutors, causing many to drop out. The team prototyped a chat-based onboarding where a short conversation elicits key preferences (subject, availability, learning style) and surfaces a starter tutor match within minutes.

To address hallucination and privacy concerns, the UI displays each extracted preference as an editable chip and highlights which statements were inferred vs. directly stated. The model is constrained to a small, auditable set of intents and slot values; fallbacks always prompt the user for confirmation before saving.

This conversational approach reduced time-to-first-lesson by 40% and improved match acceptance rate by 22% because users received personalized options quickly. Tutors also reported better initial fit and fewer reschedules. Importantly, Tutorly maintained user trust by keeping the LLM in a scaffolding role rather than an authoritative decision-maker.

For product designers, the Tutorly example shows how LLMs can compress friction if paired with explicit verification patterns and lightweight edit affordances. Designers should treat generated outputs as suggestions and always surface clear edit paths for users.