AI-Powered Onboarding: Personalizing KYC Flows Without Creepy Prompts

AI · 6 min read

The onboarding funnel for a challenger banking app was losing users on the KYC section: the long, one-size-fits-all questionnaire felt intrusive and confusing. The product team built an on-device lightweight LLM layer to rewrite and surface only the KYC fields relevant to a user's application scenario, reducing cognitive load and perceived friction.

Prompt engineering focused on brevity and tone: prompts instructed the model to convert bureaucratic fields into one-sentence clarifying questions and to identify when follow-ups were optional. Crucially, the team added privacy guardrails: the LLM received only sanitized metadata about the user's declared use-case and never handled raw identifying documents. All radical suggestions were routed through a rules engine to ensure compliance.

In an experiment, personalized KYC flows produced a 28% lower abandonment rate and a 19% faster completion time. Qualitative interviews revealed users appreciated contextual explanations that answered “why we need this” in plain language. The case highlights how small, controlled AI interventions can make regulatory UX more humane without expanding data exposure.