Manulife

In life insurance, underwriting speed directly affects conversion rates — applicants who face lengthy approval processes often abandon applications or seek coverage elsewhere. Manulife, one of Canada's largest insurers, faced this friction in its digital underwriting workflow. Its original AI-powered decision engine, Maude (Manulife Automated Underwriting Decision Engine), launched in 2018, had meaningfully improved throughput but left a substantial share of eligible applications requiring manual review. The static questionnaire design asked questions regardless of applicant context, generating unnecessary medical history data that added friction without improving risk assessment. Meanwhile, experienced underwriters were absorbed by routine cases that offered little professional development, limiting their exposure to the complex risks that demand genuine actuarial judgment.

Manulife upgraded Maude's front-end questionnaire logic using predictive ML to make the application experience adaptive. Rather than presenting a fixed set of medical questions, the updated system dynamically tailors its questionnaire based on the applicant's age, requested coverage amount, and prior responses — branching in real time to ask only clinically relevant follow-ups. Standardized medication lists and predefined response options for common conditions were introduced to feed cleaner, more consistent inputs into the underlying risk model, improving prediction accuracy without retraining the core architecture. Applications assessed as low-risk are routed through straight-through processing and approved automatically in as little as two minutes. Those involving existing medical conditions, unusual coverage levels, or ambiguous responses are escalated to human underwriters. The September 2025 rollout extended these improvements across Manulife's Canadian individual life insurance portfolio, with critical illness coverage identified as the next expansion target.

Following the September 2025 update, Maude's performance improved significantly across the board:

• Instant approval rate: rose to 58% of eligible applications — nearly double the pre-update rate
• Medical questions eliminated: up to 40% fewer questions per application, reducing applicant drop-off
• Approval speed: qualifying applications decided in as little as 2 minutes

Qualitatively, human underwriters shifted almost entirely to complex cases, improving both caseload quality and professional development opportunities for newer staff. The adaptive questionnaire also reduced applicant friction, supporting higher completion rates on digital channels where abandonment is a persistent industry challenge.

• Questionnaire design is a model input problem: improving the quality and relevance of data collected — not just the model itself — can dramatically increase straight-through processing rates.
• Contextual, adaptive questioning reduces friction: tailoring questions to applicant profile removes irrelevant items, cutting completion time and improving data consistency for the ML model.
• Graduated automation builds institutional trust: starting with low-risk, clearly-bounded cases and expanding scope incrementally mirrors actuarial best practice and reduces the risk of model failures at scale.
• Automation elevates human roles rather than eliminating them: routing complex cases to underwriters improves job quality and preserves the expert judgment that automated systems cannot yet replicate.