AI isn’t replacing insurance, it’s finally making it work

Why the next wave in protection will be human-led, machine-scaled and built across Asia’s thriving insurance hubs.

Insurance has always promised speed, certainty and care, then asked customers to accept friction, opacity and waiting. 

The last two years suggest that the gap is finally closing. Artificial intelligence (AI) and intelligent automation are moving from pilot to production in underwriting, claims, and distribution.

The result is not a robot insurer, but a re-balanced model: humans for judgement and empathy, machines for the heavy lifting at scale. From our vantage point in Singapore and Hong Kong, this shift is most visible in Asia, where insurance penetration still lags needs but digital adoption is high. 

Consider three forces making AI decisive now.

First, the need is acute. Across 12 surveyed Asian markets, Swiss Re estimates a 2024 health protection gap of $258b in premium-equivalent terms, up 21% since 2017. That is a measure of healthcare costs households find financially stressful, and a reminder that efficiency is not a nice-to-have, it is a public good.

Protection shortfalls are not limited to health. Natural catastrophe losses remain persistently underinsured worldwide, a gap Swiss Re tracks region by region, with Asia a frequent hotspot.

Second, the technology works. In the UK, Aviva reports more than 80 AI models in its claims domain, trimming liability assessment time for complex cases by 23 days, improving claim routing accuracy by 30% and reducing customer complaints by 65%. This resulted in savings of over $79m (£60m) in 2024. 

Those are not proof-of-concept, they are P&L outcomes. In distribution, embedded insurance — protection offered seamlessly inside everyday transactions — is scaling fast. Market data sized global embedded premiums at about $119b last year, and projects a near-28% CAGR to 2032, with Asia Pacific already contributing 46% of the market.

Third, the regulatory scaffolding is catching up. Singapore’s Monetary Authority has updated its AI governance for financial services with model risk management guidance that makes fairness, accountability and transparency concrete expectations for regulated firms, not slogans. That clarity is catalysing responsible deployment across the region’s two great insurance gateways, Singapore and Hong Kong.

The commercial momentum is visible. Hong Kong posted record new insurance premiums in the first half of 2025, up 39% year-on-year to $12.7b (HK$99b), buoyed by sophisticated cross-border demand. 

Regional carriers are retooling: Prudential’s 2025 growth narrative highlighted Hong Kong and mainland China as engines of new-business profit, with Indonesia and Singapore expanding via broader channels, a picture of distribution modernisation in action.

And in Singapore, consolidation and foreign investment point to scale seeking technology leverage. From our own insurance business’s perspective, three design principles should guide how AI reshapes the industry without hollowing out its human core.

But first, allow me to share some of our proof points to set the scene for why these principles are so powerful. 

There is ample evidence that this works when done well. Leaders deploying AI in claims and service show measurable improvements in cycle time and satisfaction, a pattern that should now be the baseline, not the outlier. 

Build for distribution as an ecosystem, not a channel. Asia’s protection gap will not close through captive agents alone. The most efficient model is B2B2C: cloud APIs that let banks, platforms and affinity partners embed regulated advice and instant binding into the flows customers already use. 

Done right, embedded is not a pop-up add-on. It is context-aware cover with clear disclosures and claims that are as digital as the sale. In our own operations, we prioritise a capital-light, partner-first approach to help incumbents digitise quickly rather than reinvent the stack. The market tailwinds are strong, with Asia already the largest share of embedded premiums globally.

Treat AI governance as product design, not compliance theatre. Trust is a design choice. In practice, that means aligning model monitoring, data lineage and outcome testing with local expectations, and pairing digital rails with licensed advisory where appropriate. 

Singapore’s updated AI model risk guidelines are useful precisely because they convert abstract principles into operational controls. Boards and product teams can build them.

What does good look like on the ground? In our group, we have invested in AI-assisted consultation for discovery, and claims automation for relief. The north star is a customer experience that feels human-fast: instant for the routine, empathetic for the complex. We use AI to reach the customer, and the human touch to keep faith with them. 

The Asia advantage
Asia’s insurance hubs are unusually well placed to lead this transition. Digital demand meets regulatory clarity. With MAS licensing and oversight, Singapore has become a safe place to test AI-enabled advisory and servicing at scale across ASEAN. 

Meanwhile, Hong Kong’s cross-border flows are re-energising savings and protection products that increasingly rely on digital onboarding and remote servicing. Growth markets to learn fast. Vietnam’s insurtech scene shows how quickly embedded and comparison platforms can scale when paired with modern claims and product rails. 

There is a clear economic case for speed. The region’s insurers posted solid embedded-value gains last year, with Vietnam up 25%, Malaysia 13.3% and Taiwan 13.7%. This provided both the balance-sheet cover and the strategic incentive for bolder digital moves.

Where the industry must be honest
This is not a straight-line technology story. AI can amplify poor incentives as easily as good ones. Three risks require candour: opacity, data deserts and kitchen-sink pilots.

Black-box underwriting erodes trust. Explainability must be adequate for regulators and meaningful for customers. Singapore’s FEAT principles were early to state the goal, but model risk management now makes it enforceable.

Many Asian markets have sparse, noisy claims data. The answer is not optimism, it is consortium-grade data sharing and synthetic augmentation governed transparently. Generative AI can feel like a Swiss Army knife, but insurers need fewer, bigger bets tied to measurable journey outcomes: days saved in claims, percentage-point gains in persistence, complaint rates down.

The reward for getting this right is not incremental. McKinsey estimates generative AI’s total economic potential in the trillions. The carriers capturing it first will be those that re-platform processes, not just bolt on models.

A humane way to insure
For decades, the industry has asked customers to wait while we verified, reconciled and reviewed. AI lets us keep the promise we made at the point of sale: that when life happens, help arrives fast, fair and comprehensible. 

The technology is finally capable, the regulatory guardrails are in place, and the demand is undeniable from Hong Kong to Singapore and Ho Chi Minh City.

Our stance is straightforward. Use AI to compress the distance between worry and resolution, keep people where trust and nuance live. Technology should make life easier, not louder. If we build to that standard, insurance can move from paperwork to protection.