Manulife's underwriting engine can approve a life insurance application on its own and cannot decline one. Every decline routes to a human underwriter. The Manulife Automated Underwriting Decision Engine, MAUDE, is deliberately one-directional.
That is an architecture, not a limitation being worked around, and it decides which of the two underwriting failure modes the carrier keeps. Manulife plans 200 additional AI systems in 2026 against a $1 billion enterprise value target by 2027.
Key Takeaways
- Approve yes, never decline. The engine issues approvals without a human touching the file and forwards everything it cannot clear, so the algorithm owns one side of the distribution outright.
- More than 58% of eligible advisor-submitted cases auto-approve, with notification in two minutes, a 56% increase in the auto-approval rate after the latest upgrade.
- Running since 2018, when it launched in Canada as AIDA, so the earliest automated cohorts already carry several years of developing claims experience.
- 200 additional AI systems in 2026 and a $1 billion enterprise value target by 2027, which is a scale that requires touching a meaningful share of new business rather than a single product line.
- 59% of individual life applications are already no-exam decisions industry-wide, the category the NAIC's accelerated underwriting guidance covers in 2026 market conduct examinations.
An Engine Built to Fail in Only One Direction
MAUDE screens an application and, where the risk profile clears its thresholds, issues an approval with no underwriter involved. Where it does not clear, the file goes to a person rather than to an automated decline. The decision space is asymmetric by design.
This is not a new deployment. The system launched in Canada in 2018 as AIDA, the Artificial Intelligence Decision Algorithm, the first tool in the country to make automatic life underwriting decisions. By the most recent reported period it was auto-approving more than 58% of eligible advisor-submitted cases with notification inside two minutes, a 56% increase in the auto-approval rate from before the upgrade.
Around that core, Manulife has deployed AI sales enablement across nine markets and selected Akka to run the enterprise platform coordinating underwriting, distribution and operations. Evident AI ranks the company first among life insurers globally for AI maturity and responsible innovation.
The eight-year lineage matters more than the ranking. An approve-only engine running since 2018 has real claims experience developing behind its earliest automated cohorts, which is exactly the retrospective evidence a newly launched system cannot offer.
The Asymmetry Moves the Risk, It Does Not Remove It
An underwriting model that can both approve and decline carries two failure modes. A false decline wrongly rejects an insurable life, and that is the one that draws an unfair trade practices complaint or a disparate impact review. A false approval accepts a life that should have been rated or declined, and that one works on the mortality assumptions inside the pricing.
An approve-only design eliminates the first by construction. It does not touch the second. Every approval MAUDE issues is one the underwriting department accepts without a second look, so the whole of the model's risk sits in where the accept threshold is set.
The observability is asymmetric too, and that is the sharper way to put it. The triage call, deciding which files are clean enough to clear, is itself a model output rather than a simple rule. Miscalibrate it toward forwarding too many approvable cases and underwriters drown in work the system should have handled, which someone notices within a week. Miscalibrate it toward clearing marginal cases and nothing registers at all, because an approval the engine issues alone generates no workload signal.
That is where the pricing consequence lands. The engine's calibration is only ever tested against the population it is willing to approve, and there is no decline-side comparison inside the automated book to benchmark it against. If the threshold drifts even slightly permissive under pressure to sustain a 58% approval rate, the anti-selection accumulates entirely in files no person read, and it surfaces when claims develop rather than when the drift occurs.
The industry is moving the other way. Deloitte's 2026 outlook reports early agentic deployments delivering 36% underwriting efficiency gains and 40% reductions in claims cycle time from agents holding end-to-end authority, and Celent finds 22% of insurers planning agentic deployment by the end of 2026, rising toward 70% by 2028. Manulife's boundary is the conservative case in that field.
Examining a Population Nobody Read
The NAIC's accelerated underwriting guidance, issued in August 2024, takes effect in 2026 market conduct examinations and covers predictive models, external data sources and no-exam decisions. That is precisely the category MAUDE's approval path occupies, in a market where no-exam decisions already handle 59% of individual life applications.
For the decline side the framework fits cleanly. Every actual decline carries a human underwriter's judgment and a file an examiner already knows how to read.
The accept side is where the framework has no established method. Reviewing a book of approvals that no person individually assessed cannot be done file by file, because there is no underwriting judgment in any file to review. It requires aggregate outcome monitoring across the automated population, which is a different evidentiary product from the one market conduct examination was built around, and neither the guidance nor Manulife's public disclosures describe what that evidence looks like.
The commercial pressure runs against the boundary as well. A $1 billion enterprise value target spanning underwriting, distribution and operations, on top of 200 additional systems in 2026, is a number that needs volume. The design's discipline is that the algorithm never says no. Whether it stays one-directional is a question the target and the architecture answer differently.
Further Reading
- NAIC's Accelerated Underwriting Guidance Takes Effect in 2026 – the market-conduct examination framework now covering predictive models and no-exam decisions in individual life underwriting.
- Mortality Slippage Tests AI Life Underwriting at Scale – why industry-average 15% mortality slippage in accelerated underwriting programs takes three to five years of claims data to validate, the exact evidence gap MAUDE's approve-only design has been closing since 2018.
- How Actuaries Validate AI Models for State Rate Filings – the broader model validation workflow an approve-only underwriting engine would need to satisfy.
- The AI Governance Gap in Actuarial Practice – why professional standards have not kept pace with production-scale agentic underwriting deployment.
- Predictive Analytics in Insurance Underwriting 2026 – the broader adoption context of GLMs, gradient boosting, and agentic platforms across life, health, and P&C underwriting.
Sources
- Insurance Innovation Reporter, "Manulife Sets Course to Become an AI-Powered Insurer"
- The Logic, "Manulife now uses AI to say yes to life insurance applications"
- Manulife Canada, "Manulife Canada Delivers Faster Life Insurance Approvals with AI Integration and Enhanced Digital Application"
- Investment Executive, "Manulife updates underwriting AI, doubles instant approval rate"
- AI for Insurance, "Manulife upgrades AI underwriting engine MAUDE to accelerate life insurance decisions"