Diana Bui speaks with industry experts to test whether AI’s growing momentum in captive insurance lives up to the hype
Momentum or myth?
Artificial intelligence (AI) is rapidly reshaping industries worldwide, and captive insurance is no exception.
Routine tasks such as data extraction, loss-run cleansing and reserve calculations have left the desks of overstretched analysts and now sit with unfailingly consistent algorithms.
The benefit extends far beyond speed. With modern systems capable of analysing millions of records almost in real time, captive managers now have a sharper view of volatility, pricing and capital efficiency than traditional techniques ever allowed.
Jonathan York, chief technology officer at Luzern Risk, sees the change first-hand even in a firm born digital.
“AI is transforming the captive-management business in significant ways,” he says. “What started as a few efficiency gains is fast becoming a fundamentally different way of working.”
The momentum is just as evident for captive brokers and managers. Queena Cheung, chief strategy and digital officer at Marsh Captive Solutions, observes: “In captives, we launched the first and only AI-powered captive product, ReadyCell, which enables clients to open a cell captive within minutes instead of weeks or sometimes months.”
She adds that AI and GenAI “greatly enhances risk analysis and underwriting by enabling data analysis at greater speeds and complexities,” while a deliberate ‘human-in-the-loop’ approach helps keep bias in check.
AI is also levelling the field for smaller employers. “Self-funded or captive insurance has traditionally been complex and required a lot of resources to quote and administer,” says Tawfiq Bajjali, chief technology and product officer at ClearPoint Health.
“AI-driven digital platforms, which include machine learning and advanced-analytics algorithms, can rapidly analyse a small and medium-sized enterprise’s (SME’s) insurance and medical data, making it easier to evaluate and underwrite alternative funding options and manage policies throughout a policy lifecycle.”
Global insurers have responded by opening the throttle on internal programmes. Ashok Krishnan, chief innovation, data and analytics officer at AXA XL, calls AI “a core part” of the firm’s strategy.
“We rolled out Secure GPT, which taps the power of generative-AI large-language models and makes it available to all employees. Today, all 10,000 colleagues in our 25–26 countries have access to this technology.”
AXA XL is also “one of the first insurers to test Microsoft Copilot in beta. About 500 employees are using it right now,” Krishnan says. Beyond those democratised tools, Krishnan points to “targeted solutions for underwriters and claims handlers — tools designed to make their work easier and deliver more value to clients.”
Risk assessment in real time
If the first wave of adoption replaced manual toil, the second is about shrinking the distance between signal and decision. At Luzern Risk that shortening gap is already visible.
“Today, AI helps us extract data from complex documents, interact with sophisticated models using natural language, and streamline underwriting processes,” says York.
The automation, he adds, “frees our team to focus on high-value tasks like designing more tailored products, managing difficult client situations with greater nuance, and identifying and solving problems more effectively.”
York is convinced the gains will compound. “Those companies that successfully integrate these technologies will become dramatically more efficient; not just 10 per cent better, but multiple times more productive in delivering high-quality services to clients.”
The same immediacy is taking hold at Marsh Captive Solutions. Cheung points to Sentrisk, “a cutting-edge AI-powered platform that analyses best-in-market data, empowering businesses with a new way to manage global supply-chain risk.”
The system taps shipping manifests and geospatial satellite imagery to map supplier networks and verify asset locations, so a blocked canal or factory fire surfaces on dashboards almost at the moment it happens, giving captives time to adjust treaties before losses accumulate.
Internally, Marsh runs a suite of GenAI capabilities: “We use a variety of internally developed and externally sourced AI tools, including text-extraction automation, document summarisation, chatbots and other customer-service interactions. Many of our uses involve a proprietary internal generative-AI tool called LenAI.”
Notably, Marsh does not train LenAI on user inputs, and LenAI does not retain any outputs it generates.
Apart from using LenAI for everyday tasks, Marsh also utilises systems such as Sentrisk, “a cutting-edge AI-powered platform that analyses best-in-market data, empowering businesses with a new way to manage global supply-chain risk,” and analytics platforms like Blue[i] and Data Navigator parse claims trends so risk managers and captive managers can together see where programme tweaks would have the biggest impact.
The system taps shipping manifests and geospatial satellite imagery to map supplier networks and verify asset locations, so a blocked canal or factory fire surfaces on dashboards almost at the moment it happens, giving companies time to adjust treaties before losses accumulate.
For Bajjali, speed is equally critical in employee benefits. ClearPoint’s system, he says, “extracts data from plan documents, claims and demographic files and then matches each employer to an optimal alternative funding type — group captive, traditional stop-loss or level-funded — based on that risk analysis.
“From there, we review medical history and classify risk to be sure the right cost-management solutions are built into the policy and that the risk is matched to the right carrier.” The pay-off, he says, is immediate: “Doing so saves underwriters time in reviewing each case and speeds up their decision-making.”
Buying cover is quicker too. “We use AI to make quote comparison easy. Instead of manually sorting and reformatting multiple quotes, we automatically summarise and highlight our recommended plan so benefit advisers and employers can compare options quickly,” Bajjali explains.
“Our approach allows benefit advisers to deliver tailored insurance options much faster and more efficiently to their SMB customers. It also means that SMBs can approach their health benefits like larger employers do.”
The momentum behind real-time, AI-powered risk assessment is undeniable. Bajjali sees the result in sharper, faster decisions, while Cheung warns that every leap in speed demands stronger guard-rails. Krishnan zooms out, calling this “day zero of the AI revolution — akin to the internet’s early 1990s.” As AI integration accelerates, the industry faces a hard question: how to harness its speed without losing sight of transparency, fairness and human oversight.
Governance, ethics and human touch
Regulators have shifted AI in insurance from a recommended best practice to a firm requirement. Under Europe’s new EU AI Act, insurers will soon be required to document data sources, demonstrate bias controls, and ensure human oversight for any “high-risk” algorithm.
The UK takes a lighter, principles-based path, while US regulators lean on the National Association of Insurance Commissioners (NAIC) bulletin alongside a patchwork of state-level rules, including Colorado’s mandatory bias testing for life underwriting.
Singapore’s Fairness, Ethics, Accountability and Transparency (FEAT) principles and Canada’s draft accountability law add further complexity. Yet the core message remains the same: innovation must be built on a clear ethical foundation.
York says Luzern Risk meets that challenge with a structured approach: “The first step is establishing clear principles to guide how we leverage AI. These principles determine which activities we use AI for and what guardrails we put in place, based on our confidence in the technology's capabilities.”
That foundation, he says, ensures transparency. “Our framework covers confidentiality, data privacy, security, and quality standards. Having these principles enables transparency with our customers, employees, and partners, and gets everyone on the same sheet of music.”
Then comes rigorous testing. “Second, we proceed cautiously — experimenting, evaluating, and back-testing across diverse scenarios, much as we would when developing human talent. This approach builds our understanding of how much we can trust each implementation.”
Finally, nothing is left uncertain. “We ensure our AI systems aren't black boxes. Every action is tracked, with automated monitoring designed to catch potential issues.” He describes it as “a continuous improvement cycle, both in development and daily operations.” That vigilance, he adds, has already freed staff to focus on “higher-value challenges like creative problem-solving and innovative structuring”.
Meanwhile, Cheung sees governance and talent development as two sides of the same coin. “The biggest challenges include data quality, systems design and colleague adaptation,” she says. To tackle these issues, Marsh has embedded an AI Risk Governance Framework to manage emerging risks, including those introduced by new legislation. An AI Centre of Excellence and a dedicated risk committee ensure the company stays aligned with evolving standards while fostering a culture of accountability around AI.
Keeping staff up to speed is also a priority. “We have dedicated captive AI experts and a multi-faceted upskilling programme, including office hours and bite-size resources, to help colleagues understand AI and become fluent GenAI users,” Cheung notes.
At the same time, the human element stays front and centre. “AI innovations enhance our colleagues’ ability to engage with clients rather than replacing human interactions,” she says. Automated note-taking allows advisers to focus on conversations, while natural-language interfaces ensure that “captive-insurance experts, who may not be data specialists, can perform analyses using natural language, tailoring complex coverage options for individual captives with more flexibility.”
Bajjali echoes that people stay on the bridge: “AI should augment, not replace, clinical judgment and data privacy is non-negotiable.” ClearPoint’s SOC 2 and HIPAA-compliant cloud encrypts everything, and role-based controls keep data exposure “minimal and necessary”.
To help newcomers, the firm built an alternative-funding marketplace plus a Centre of Excellence: “These offerings allow our partners to participate seamlessly and meet them where they are.”
Krishnan sees ethics as both a shield and a spear. “Any technology opens good and bad possibilities; our job is to ensure the good far outweighs the bad,” he says. At AXA XL, every AI model undergoes an ethics review tailored to local regulations.
He highlights the need for every organisation to ensure compliance with regulations. “The EU AI Act is already live, with more provisions taking effect in the coming months,” he notes, adding that AXA XL is “fully compliant and tracking similar regulations in every region where we operate.”
Keeping people ahead of the curve is critical. “You won’t lose your job to AI, but you could lose it to someone who knows how to use AI better than you,” Krishnan highlights. To address this, continuous training runs alongside model testing, ensuring employees and clients feel empowered rather than replaced.
Mapping the road ahead
Ask the industry’s technology chiefs what comes next and they paint a future in which algorithms and people work side by side, each amplifying the other’s strengths. York sees the proof of concept already.
“It is still early, but several AI applications are showing particular promise. We are retrieving data from vast stores of documents, summarising key passages in seconds and producing first-draft analyses for stakeholder review,” he says.
Those time savings, he argues, are not being banked as idle minutes. “The efficiencies free our team to focus on higher-value challenges like creative problem-solving and innovative structuring.”
Luzern now runs controlled pilots for every new model, collecting evidence before rolling it into live production.
York is convinced that the most dramatic gains will come once AI agents graduate from back-office helpers to full collaborators.
“We expect these systems to spot opportunities to improve captive health, respond to regulators more quickly and test a wider range of scenarios. How the evolution plays out is not yet clear, but our commitment to reliable, high-quality service will not change.”
At Marsh Captive Solutions, Cheung tracks a similar trajectory. “Over the next 5 to 10 years, AI-driven technologies are likely to lead to a more data-driven and proactive approach to risk management,” she says.
The aim is to let risk managers “respond more proactively to the needs of businesses in a rapidly changing landscape”, she says, turning captives into dynamic risk platforms rather than static financing vehicles.
Health-benefits captives have an even longer runway, argues Bajjali. Many advisers and small employers are only now encountering alternative funding for the first time, so ClearPoint has built what Bajjali calls “a tech and intelligence-powered marketplace” around them.
“We developed a curriculum, hired field teams who understand captives and offer outsourcing through our Centre of Excellence. Partners can join the market at their own pace,” he says. Employees, too, will feel the difference.
“In our model staff are not passive recipients of a plan; they engage with AI insights that help them navigate benefits and care. Coverage becomes personal, outcomes improve and costs fall.”
Bajjali expects the data floodgates to open within a decade. “Data siloes in healthcare and insurance will break down, and we’ll see a lot more interoperability of health data from providers, employers, insurance carriers, digital mobile apps, devices, and pharmaceutical companies.
“This will significantly improve the AI algorithms in play allowing for more personalisation and successful care interventions that lower costs and improve health outcomes.”
ClearPoint’s CliniCaptive project is already working with hospital groups willing to be paid on value delivered rather than services rendered. He predicts the result will be “more stable and highly cost-efficient captives”.
Meanwhile, Krishnan believes AI will redraw the boundary between insurable and uninsurable. “Risks once deemed uninsurable — flood-prone properties, wildfire zones, complex cyber exposures — can be reassessed with better modelling,” he says.
“AI cannot stop earthquakes or hurricanes, but it can improve prediction and prevention. Once we understand a risk precisely, we can design solutions that were impossible before.”
Krishnan believes AI is still at a very early stage but will soon be inseparable from the way the industry operates. “Over the next 5 to 10 years, AI will permeate everything, acting not as a standalone tool but as an enabler that helps us do our work faster and better,” he says. For AXA XL the destination is clear: smarter underwriting, more responsive claims and a broader social benefit. “The challenges are real, yet the transformative potential is greater,” he concludes.
Across all the discussions a clear consensus emerges — captive managers are embracing AI not just as a faster way to handle routine work, but as a tool that can pull hidden insights from complex datasets, retrieve key information on demand, and structure complex material into decisions that matter. Yet no one underestimates the need for balance.
As York reflects, the future will be shaped by the collaboration between human expertise and machine intelligence, and whatever form that partnership takes, the commitment to delivering high-quality, reliable client service will remain unchanged.
