Manulife

Life insurance underwriting at Manulife required human agents to manually review every incoming application — a process that took days and created a bottleneck that constrained sales capacity. The legacy rules-engine system that preceded AI review was fundamentally rigid: each decision factor (a specific medication, health condition, or age threshold) had to be programmed individually, making the system binary and slow to adapt. As medical treatment patterns evolved — such as widespread shifts from one drug class to another — the rules engine required manual updates to stay current. At scale, this meant the firm could only process as many applications as its human underwriting staff could handle, capping growth and leaving low-risk applicants waiting unnecessarily for straightforward approvals.

Manulife developed MAUDE (Manulife Automated Underwriting Decision Engine), a predictive ML model trained on historical underwriting decisions made by human agents, capturing how risk was assessed across medical history, age, and other applicant factors. An early version launched in 2018, with a substantially upgraded version rolled out to client-facing advisors in fall 2025. MAUDE performs the first review of all eligible applications, automatically approving low-risk applicants within minutes and routing higher-risk cases to human underwriters. Crucially, the system is scoped to approvals only — it cannot deny coverage, a deliberate governance boundary. Manulife also redesigned its application forms to begin with current medications and dynamically surface linked conditions, feeding cleaner, more structured data into the model.

MAUDE now automatically approves 58% of eligible applications, having signed off on tens of thousands of policies since deployment. The most direct impact is speed: approvals that previously took days are now completed within minutes for low-risk applicants. Qualitative outcomes include increased underwriting throughput, expanded sales capacity, and a simplified application experience for policyholders. The system also frees human underwriters to focus on complex, edge-case applications where judgment and nuance are most valuable. MAUDE's performance contributes toward Manulife's stated goal of generating $1 billion in enterprise value from AI by 2027, one-fifth of which is targeted through efficiency gains.

• Scope AI to approvals, not denials. Restricting MAUDE to saying yes — never no — manages regulatory and reputational risk while still capturing the majority of efficiency gains, since most applications are low-risk.
• ML models materially outperform rules engines in dynamic domains. Unlike binary rules systems, predictive models infer decisions from all available data simultaneously and adapt more readily when treatment patterns shift (e.g., medication substitutions across a patient population).
• Application redesign amplifies model performance. Restructuring forms to surface structured, medication-first data improved input quality, not just processing speed.
• Bias auditing is non-negotiable in insurance AI. Manulife runs post-launch audits with cross-functional teams to verify decision correctness — a required practice given regulatory scrutiny of AI in federally regulated financial institutions.
• Plan for false positives as a cost of scale. Misjudged approvals are priced into policy costs, treating AI error rates the same as human underwriter error rates.