Who this is for: AI and content products where users churn at the front door because the product asks them to choose before it has earned the right to ask anything.

Learntech had strong content and a 31% course completion rate. Users landed on a catalogue of 200+ options, froze, and left. The instinct internally was to cut the catalogue. The actual problem was that the product opened with a decision the user was not equipped to make.

Reframed onboarding as a conversation rather than a filter. The AI asks about goals, available time and current level in plain language, then assembles a path - so the user's first action is answering a question about themselves, not evaluating 200 unknowns.

The hard design constraint was tone. Early prototypes felt like an intake form: too many questions, too clinical. I cut the flow to the smallest number of questions that could produce a defensibly personalized path, and designed the AI's responses to reflect back what it had understood at each step so the user could feel the thing working.

Designed the recovery states explicitly - what happens when the AI misreads someone, and how a user corrects their path without starting over. This is where most conversational onboarding quietly fails.

Built a full interactive prototype in Framer and validated it with real users before engineering wrote any code, so the team was not debugging conversational tone in production.

Two months, embedded with the product and engineering leads. Prototype-first: every flow was tested in Framer with real users, and I brought recordings of people getting confused to the team rather than opinions. Two rounds of validation before handoff, then paired with engineering through build.

Figures reported by the Learntech product team over the six months following launch.

Onboarding drop-off down 25%

Course completion improved from 31% to 64%

Approximately 10,000 new subscribers over the following six months

Average session length up 4.7 minutes

User satisfaction reported at 4.7 / 5

Referral coefficient reported at 0.4

Figma

Framer

Conversational AI patterns

Prototype validation

AI onboarding design for a product with a front-door churn problem. Typically 3–5 weeks: current-state audit, conversational flow design, tested Framer prototype, and a handoff spec including failure and correction states.

Typical investment: $4,500 – $7,500. Standalone onboarding audit with recommendations: $1,500.