Using generative AI to personalize onboarding copy at a fintech startup — before and after

AI · 6 min read

NovaLedger, a mid-stage fintech, had stagnant onboarding completion and high drop-off on identity verification screens. The design team integrated a GenAI layer that produced contextual microcopy and step-specific clarifications based on a few deterministic user signals: user segment, stated goal, and verification progress. Instead of full AI-driven flows, designers used the generator to produce variants that were reviewed and curated into a content bank.

The implementation emphasized guardrails: all generated copy passed through a rules engine that removed regulatory-sensitive phrasing and ensured consistency with legal scripts. Designers created prompt templates and a small review dashboard so product managers could see which generated variants performed best. Importantly, the team measured lift with randomized rollout and a strict metric hierarchy focusing on completion, error-rate in uploaded documents, and time-to-verify.

Results were notable: a 14% relative increase in onboarding completion and a 9% reduction in support tickets for identity-related issues. The after-state was not full automation but a human-in-the-loop system where designers curate high-performing AI variants, giving them scale without losing control. The experiment demonstrates how GenAI is most effective when used to expand the design toolkit under operational constraints rather than as a replacement for design judgement.