AI is everywhere now in the insurance industry and yet talk of true ‘digital transformation’ seems stuck.
If this transformation is about efficiencies, automation, and working around legacy mainframes, then yes, the use of artificial intelligence is part of an ongoing process – perhaps speeding things up, but not breaking new ground.
If it’s about fundamentally reimagining business processes, customer engagement and product innovation, then perhaps talk of AI, particularly generative AI and the use of large-language models, is better for conference fodder. To be sure, no one in the industry seems to be rethinking incentives of how insurers work with banks and agency forces to sell products.
That is the impression DigFin has from the public comments at a recent industry event on digital insurance.
Big threat, big play
On the one hand, there is a sense of drama. The intuitive, democratic nature of genAI has rightfully put the scare into C-suites.
As Orchis Li, General Manager at Gen Re Hong Kong, said, “It’s no longer about digital transformation, it’s now about survival.”
From the perspective of efficiency gains, firms are now enthusiastically embracing automation tools. While this is not new, genAI brings this process down to individual initiatives, instead of just being about a top-down project managed by the chief technology officer. This is exciting.
Most firms have already gone through digitization, converting analog records into structured, machine-readable data, and using it to ditch manual processes. From there, firms can begin automate parts of decision-making by integrating them with data flows.
Jim Qin, CEO of Zurich Hong Kong, says his organization moved from 25 percent to over 90 percent digitalization in just a few years, creating a foundation for data-driven decision-making. “Every piece of data is digital and structured, so we can use it,” Qin said.
The next step: leveraging AI to unlock the full potential of this data, automating claims, underwriting, and customer profiling, and ultimately striving for an “AI-native” enterprise.
Or something more ordinary
The rhetoric (AI native?) is frothy, though.
Yes, many parts of the industry are perfect for automation, and firms are deploying predictive modeling for underwriting and claims, as well as more prosaic things such as processing documents, generating reports, automating presentations, and interacting with customers. More employees are going to be able to use AI tools without deep technical expertise, which further spreads use throughout an organization.
But these all make existing models work better. They are not sea changes, just iterations. There’s nothing wrong with incremental improvements, but is it “digital transformation”? Are insurers “AI native”?
David Piesse of the International Insurance Society and an experienced investor and advisor in insurance-related tech companies, says fundamental changes won’t come until the industry can use agentic AI, systems that combine large language models with causal reasoning and human knowledge.
The key word is ‘causal’. The current AI industry is built on massive compute to identify correlations from big data. Throw together enough sets of data, and an AI can tease out patterns that a human would struggle to find. But the AIs don’t know why X leads to Y, or if X+Y are what enable Z, or if it’s something else. They just know that when X is there, conditioned by Y, we get Z, but that may be a coincidence.
This might be acceptable to a business that just wants to achieve a certain outcome. No one cares if it’s causation or correlation. But the advent of genAI based on big-data models has taught everyone the hazards of correlation: these models make mistakes. They even make shit up, because that’s what the coincidental patterns (or flawed data inputs) suggest.
Digital transformation is about far more than using genAI to come up with a marketing document. This may be why Piesse says the industry has “overdosed” on genAI. Real transformation would mean using technology to rethink business models, enabling ideas that are still mostly fintech dreams such as parametric insurance.
Barriers to bolder
To do so will likely mean escaping the bonds of legacy technology stacks.
Legacy systems, some dating back decades, are a persistent drag on innovation. Migrating from mainframes to modern, cloud-based architectures is a multi-year, sometimes decade-long, endeavor.
Data integration is another stumbling block. Many insurers struggle to build even basic data lakes, let alone the sophisticated warehouses needed for AI-driven insights.
Neo Lin of Alibaba Cloud International observes that while banks have largely mastered data consolidation, insurers often lag, hampered by fragmented systems and inconsistent data standards.
Gary Ho, CIO of AXA Hong Kong and Macau, said, “Talking about AI is useless without quality of data.” The goal is not just to make internal processes more efficient, but to deliver tangible benefits to customers and distribution partners, streamlining everything from policy servicing to claims.
Data integrity is also a security issue, as it concerns ensuring information is accurate, traceable and protected from tampering. These issues will become more prominent as insurers rely more on AI-driven decision-making, even if they say they have a ‘human in the loop’. This is a new vulnerability, different to widely known management of encryption and network security.
Regulation adds another layer of complexity. Selina Lau, CEO of the Hong Kong Federation of Insurers, notes that risk-based capital and new accounting standards consume resources and can slow innovation. Legal and compliance teams wield significant influence, and without their buy-in, even CEO-backed initiatives can stall. “If the CEO says yes but legal/compliance says no, it’s a no-go, because we are a heavily regulated industry,” Lau said.
Meaningful iterations
This doesn’t mean insurers are not putting AI to useful work.
Automation promises dramatic efficiency gains, freeing up resources for growth. For example, Michael Shin, CEO of RGA Korea, highlights how AI-powered OCR has streamlined medical underwriting, reducing errors and turnaround times. Machine learning models are now used to predict claimants’ needs and costs, allowing for more dynamic product offerings.
AI also opens doors to new products and markets. In healthcare, for example, advanced analytics can identify early signs of diseases like Alzheimer’s, enabling insurers to offer targeted products for early intervention—an approach that would be impossible without the computational power and insight AI provides.
Embedded insurance and insurance-linked securities are further examples. These models rely on accurate, real-time data and sophisticated risk assessment, areas where AI excels. By enabling more precise pricing and risk selection, insurers can offer microinsurance and parametric products tailored to underserved populations, expanding their reach and impact. These are nascent, as they rely on reliable, captive sources of data, but point towards truly innovative businesses.
Perhaps the current phase in AI represents the best chance to escape the “technical debt” of still relying on mainframe computers and dead computing languages. After all, people can use AI to recode a lot of the legacy software.
The pace of change is accelerating. Maxim Afanasyev, head of financial services at Google Cloud, observes that 37% of their financial institution clients already have generative AI in production, with another 48% either recently adopting or planning to do so. In insurance, the adoption rate is even higher, reflecting the sector’s recognition that AI is now a competitive necessity. We can expect to keep hearing all about it at the next conference.
