AI-Powered Onboarding: A Case Study with Ledge Finance

AI · 6 min read

Ledge Finance, a B2C investment app, experimented with an AI onboarding assistant to tackle drop-off during the first session. They implemented a small retrieval-augmented generation (RAG) model that pulled user-specific eligibility rules, terminology, and contextual help from a curated knowledge base, then surfaced concise next-step suggestions during account setup.

Designers used conversation maps and 5 usability test rounds to craft a 3-turn flow: clarify intent, confirm constraints, and recommend the minimum viable setup. Engineers containerized the model inference and applied strict redaction rules on PII; a human-in-the-loop layer handled ambiguous cases flagged by a confidence threshold below 0.6. Privacy and regulatory teams required logs to be pseudonymized and model prompts to exclude raw account numbers.

After rollout to 15% of new users, Ledge saw time-to-first-investment fall by 31% and activation rate (first trade within 7 days) rise from 12% to 16.5%. Costs were non-trivial: API inference fees and moderation overhead increased CAC slightly, and the product team invested in monitoring for hallucinations and drift. The case shows that AI can accelerate product adoption when combined with conservative guardrails and a narrow, testable scope.