AI Application in Reinsurance: The Three Things 33 Practitioners Took Away From the SCI Course

Earlier this year I delivered 'AI Application in Reinsurance' for Singapore College of Insurance -- the first time I ran this course entirely over Zoom. Thirty-three participants joined from across the reinsurance and insurance sectors. The format was virtual; the quality of questions was not.

What follows is a distillation of the three ideas that drove the most discussion during the session, and why I think each one will matter more in 2028 than it does today.

1. AI Is Not a Productivity Tool -- It Is an Ecosystem Transformation Platform

Most conversations about AI in reinsurance start and end with efficiency: faster document processing, quicker catastrophe model runs, automated treaty checking. That is useful. But it is not the opportunity.

The distinction I drew in the session is between progressive innovation and disruptive innovation. Progressive innovation makes existing processes faster and cheaper. It is valuable and worth pursuing. Disruptive innovation redesigns the underlying model entirely.

In reinsurance, disruptive AI applications look like real-time risk exchange platforms that match cedents and reinsurers dynamically based on live exposure data. They look like parametric 2.0 products that trigger automatically on AI-verified event parameters without loss adjustment delays. They look like AI-native managing general agents that underwrite entirely in-model, with human review reserved for genuine edge cases.

“The fast fish eats the slow fish, not the big fish eating the small fish. Scale no longer guarantees survival. Speed of adaptation does.”

The reinsurers and brokers who treat AI as a cost reduction exercise will achieve incremental gains. The ones who treat it as a redesign opportunity will define the next market structure.

2. Governance Must Scale With Ambition

The session spent significant time on the governance architecture that AI adoption in reinsurance requires. The core framework I use is what I call the Traffic Light Model.

• ◆Green: AI automates the decision fully. Output is acted on without human review. Applies to routine, high-volume, low-stakes decisions where AI performance is validated and monitored continuously.
• ◆Amber: AI generates a recommendation and a human reviews before acting. Applies to material underwriting decisions, complex treaty structures, and situations where regulatory accountability requires a named decision-maker.
• ◆Red: Human decision only. AI may provide data and analysis but the decision itself must be made and documented by a qualified professional. Applies to novel risk categories, significant financial commitments, and cases where explainability to a regulator or cedent is required.

Two governance requirements that I regard as non-negotiable: override rights must be designed into AI systems from the start, not bolted on as an afterthought when a regulator or client asks for them. And every AI use case deployed in a reinsurance operation must have a pre-defined stop condition -- the specific trigger that causes the system to pause and escalate to human review.

Without a stop condition, an AI system operating at Green will keep running through edge cases it was not designed for. The stop condition is the governance equivalent of a circuit breaker.

3. The Best Reinsurance Professionals of 2028 Will Direct, Challenge, and Govern AI

The question I hear most often from practitioners is whether AI will replace reinsurance underwriters, actuaries, and brokers. My answer is consistent: AI will not replace reinsurance professionals. But reinsurance professionals who cannot engage substantively with AI will be outperformed by those who can.

The skill that will differentiate senior reinsurance professionals by 2028 is not the ability to build AI models -- that is an engineering task. It is the ability to direct what a model should optimise for, challenge the assumptions embedded in its outputs, identify when a model is operating outside its training distribution, and govern the AI use case within the firm's risk appetite and regulatory framework.

This is a domain expert skill, not a technology skill. It requires deep knowledge of reinsurance structures, underwriting philosophy, reserving methodology, and regulatory expectations -- combined with enough AI literacy to ask the right questions of an AI output and recognise a plausible-sounding error.

The practitioners who develop this capability in the next two years will be significantly more valuable in 2028. The ones who wait until it becomes an obvious requirement will spend two years catching up.

What the Zoom Format Revealed

This was my first time delivering this course entirely virtually. I was uncertain whether a complex technical topic would sustain engagement across a full session without the dynamic of a physical room.

It did. The chat ran continuously with questions and references to participants' own AI implementation challenges. The discussion on governance frameworks was particularly active -- which suggests that the industry is moving past the question of whether to adopt AI and into the harder question of how to govern it responsibly.

Thank you to Singapore College of Insurance for the ongoing partnership, and to the 33 participants who brought real challenges into the room. The transformation does not stop at the course.

Common Questions

Related insights

About the author

Raymond Cheung is a Chartered Actuary, C-suite executive and board adviser with more than 20 years of experience across Asia in risk management, insurance, ESG and corporate governance. He is the CEO of CER Consultancy and an accredited trainer at SMU Academy and the Singapore College of Insurance.