Annual investment in data centers is projected to double from around US$500bn in 2024 to more than US$1trn as early as 2027. The investment opportunity extends far beyond server halls to electricity generation, grid infrastructure, cooling, networking, and semiconductors.

According to Allianz Research, the US and China are expected to account for around 62% of new global capacity additions through 2030, but the next wave of investment is becoming increasingly global. In Europe, Germany, the UK and Ireland remain major markets, but faster expansion is expected in Spain, Finland and Denmark, where power availability and permitting conditions can be more favorable. Across Asia Pacific, excluding China, installed capacity is projected to increase from around 9GW today to more than 28GW by 2030, with Malaysia expected to grow more than tenfold.

The sector's biggest constraints are increasingly physical rather than financial. Competitive advantage is increasingly determined by access to electricity, grid connections, permitting, specialized equipment and skilled labor. In the US alone, the construction industry faces a shortage of around 439,000 skilled workers, while an estimated 349,000 additional workers may be needed in 2026.

Climate resilience is increasingly a strategic consideration rather than an operational afterthought. Around 79% of global data center capacity is already located in areas exposed to heightened natural catastrophe risk, while 54% is exposed to chronic heat and drought stress. Some of the fastest-growing AI infrastructure markets are also among the most climate-exposed, including Northern Virginia, US, Johor in Malaysia, and Marseille, France. Acute flood, wildfire and wind exposure is highest in the Americas, affecting 86% of capacity, while chronic heat and drought stress is greatest in Asia Pacific, where 89% of capacity is exposed.

Insurance is evolving alongside the sector. As data centers assume a more critical role in infrastructure, comprehensive insurance cover has become a prerequisite for financing many large-scale AI infrastructure projects. Construction costs for a single AI campus can exceed US$20bn, with insured values rising substantially once high-performance computing equipment is installed. The global data center insurance market is projected to grow from around US$11bn today to more than US$24bn by 2030, reflecting rapid capacity expansion, rising insured values and increasing operational complexity.

Demand is expected to extend beyond traditional property cover towards integrated solutions spanning construction, engineering, property, business interruption, cyber and liability, while also creating new opportunities in areas such as energy resilience, operational continuity, and technology risk.

Allianz Commercial analysis of insurance industry data center-related claims shows that fire is the leading driver of loss severity, accounting for well over 50% of around €700mn (US$800mn) worth of losses. Natural catastrophe activity ranks second, followed by willful acts, which include crime and cyber incidents, followed by power failure. Water damage is the most frequent cause of data center claims, followed by willful acts, fire, and equipment breakdown. Business interruption is the primary driver of claims severity by line of insurance, highlighting the significant financial impact of operational downtime.

The data center risk profile is changing as facilities become larger, more complex, and increasingly connected. Hyperscale and colocation of campuses can bring together multiple tenants, construction works, servers, supporting utilities and on-site infrastructure in one physical or operational space. A single event can therefore trigger claims across property, construction, business interruption, liability, cyber, and financial lines. Real-life claims case studies show that in hyperscale facilities, damage to external cooling systems, hot works-related fire damage, and a delay in start-up caused by power disturbances have each resulted in losses in the US$50mn to US$100mn range.

To read the full report, please visit: The data center construction boom: risks and claims trends

For years, the insurance industry has talked about real-time underwriting as the goal: pull in current data, assess the risk, calculate a price, and deliver a quote almost instantly. But that definition works better for a personal auto policy than it does for a complex multimillion-dollar commercial risk. If insurers want to make real-time underwriting a reality across more of the industry, the definition needs to be broader. The goal doesn't have to be an instant decision, but rather using agentic AI and the most current information available to make better decisions, faster.

That distinction matters. Fully automated, straight-through processing can work when risks are relatively standard. But underwriting doesn't need to be automated from beginning to end to become more real-time. For complex risks, the opportunity is to automate certain areas, bring more current information into the decision, and give underwriters what they need to act faster.

Real-Time Underwriting Looks Different Across the Industry

While real-time underwriting will not look the same across every line of business, the ultimate goal can be the same: accurately assessing risk and acting on that information as quickly as possible. For standardized risks like personal auto, homeowners policies, and small commercial BOP, the result may be agentic AI solutions performing straight-through processing with human oversight. For large commercial risks, where experienced underwriters need to evaluate complex exposures, full automation may not be practical or desirable, but AI tools can expedite a number of manual processes.

Across those different types of risk, three principles can move underwriting closer to real time.

Automate the processes that can be automated. Automation should accelerate the underwriting process without removing the judgment that complex risks require. Agentic AI tools can ingest and structure submission data, identify missing information, check the risk against appetite, and summarize key information. Underwriters then review the recommendations, exceptions, and complex risk factors. By automating the work surrounding the decision, the underwriter can spend more time on the decision itself.

Incorporate more current information into risk decisions. Insurance has traditionally relied heavily on historical information. Loss histories, prior claims, and other historical data remain essential, but the past is not always an adequate predictor of what happens next.

Insurers now have access to an expanding range of external data that can provide a more current view of risk. Telematics, for example, can provide information about actual driving behavior. Climate-related risks can change the relevance of historical property loss data. A property with limited losses over the previous two decades isn't necessarily exposed to the same level of risk today. The challenge is turning that information into something underwriters can actually use by incorporating it into the models and workflows that drive decisions.

Continuously monitor risk after a policy is written. Risk doesn't stand still once a policy is bound. Events can happen during a policy term, including new claims and changes in financial conditions, which can alter the insurers' view of the risk. Rather than waiting until renewal to understand what has changed, insurers can continuously monitor meaningful developments across individual accounts and their broader portfolios.

That doesn't mean every new signal triggers changes in pricing or coverage. Instead, continuous risk assessment gives insurers a more current understanding of the risks already on their books and better information when it is time to make the next underwriting decision, including at renewal.

Putting the Real-Time Mindset Into Practice

None of this requires insurers to rebuild underwriting overnight. Three practical steps can begin moving the organization toward faster, more responsive decision-making.

1. Incorporate new data into the models that drive decisions. There's an expanding universe of external information. Beyond telematics, connected home devices and geospatial imagery provide more comprehensive information on property risk. Unstructured sources such as news and social media can provide valuable insight into litigation and financial conditions, for larger commercial risks. But more data does not automatically produce better underwriting. Carriers can determine which data meaningfully improves risk assessment and then update rating models, rules engines and underwriting workflows so those signals can influence decisions.

   All of the rating engines don't need to be overhauled at once. Insurers can start with a specific line of business or risk factor, identify an external data source that can improve the assessment of that risk and test how the new information changes pricing and underwriting outcomes. Rating logic should also be flexible enough to consume new data through APIs or other integrations rather than requiring extensive system changes each time a new source is added.

2. Orchestrate data across the enterprise. Insurers often already possess valuable information that underwriters cannot easily access. Systems such as policy administration, claims, and billing may operate separately, leaving relevant information fragmented across the organization. A significant claim could occur for a commercial account, and an underwriter might not discover it until renewal.

   This is not necessarily a data problem, but an access and orchestration problem. Connecting those workflows can make underwriting more responsive. Insurers should work toward creating a connected data layer that allows information to flow between systems and establishes consistent definitions and formats so it can be used across the organization. APIs and industry data standards can help insurers connect legacy and modern platforms without waiting for a wholesale core replacement.

3. Automate the work surrounding the decision. Even when the underwriting decision itself requires human judgment, much of the work surrounding it may not. Submission triage is a good example. Rather than having an underwriter immediately begin reviewing every incoming submission, agentic AI powered solutions can first assess whether an account meets basic criteria, identify missing information, and summarize what's been provided. If information is missing, the solution can notify the broker immediately instead of discovering the gap later.

In addition to triaging, agentic AI and automation tools can increasingly handle other manual processes including data ingestion and summarization leaving underwriters to focus on the risks and decisions that actually require their expertise.

Real-time underwriting does not have to mean instant underwriting. It means making underwriting more responsive at every stage, using agentic AI solutions to automate where it makes sense, bringing current information into the decision, and maintaining a more current view of risk after the policy is written. For some risks, that may mean straight through processing. For others, the underwriter is the decisionmaker throughout. But the opportunity across all lines is to give underwriters enhanced information sooner so they can make better decisions.

The life insurance industry faces a significant opportunity: consumer interest remains strong, but many customers struggle to understand how life insurance fits their current needs, what it costs, and how to maximize its value. As expectations evolve, insurers must move beyond traditional product-led approaches and create experiences that are relevant, personalized, and easy to navigate. 

The insurers outperforming the market have demonstrated that customer centricity is more than a business philosophy. It is a growth strategy. By reimagining how they engage consumers, empower advisors, and leverage data and AI, these organizations are strengthening customer relationships, increasing loyalty, and delivering measurable business results. 

Capgemini's latest World Life Insurance Report examines how leading insurers are closing the value perception gap and transforming consumer interest into long-term growth. The report uncovers the strategies, capabilities, and operating models that distinguish best-in-class insurers from their peers, offering a practical blueprint for organizations looking to improve engagement, reduce policy lapses, and drive sustainable performance. 

Discover how life insurers can create more relevant customer experiences, deliver trusted guidance at critical life moments, and build intelligent foundations that support personalized engagement at scale. 

Key takeaways 

• Understand why consumers value life insurance but often question its relevance and affordability.
• Learn how leading insurers increase engagement through personalized, life-stage-based interactions.
• Discover how AI-powered advisors and modern distribution models enhance customer experiences.
• Explore strategies for unifying customer data to enable proactive, intelligent engagement.
• See how customer-centric insurers achieve higher growth and lower policy lapse rates. 

Ready to learn more

Sponsored by Capgemini

Not long ago, accident reconstruction relied primarily on the physical evidence left behind after a collision. Tire marks, roadway gouges, vehicle damage, scene measurements and witness statements formed the foundation of most investigations.

Those elements remain essential today. The difference is that they are now joined by an ever-growing collection of digital evidence that can provide additional insight into how a crash unfolded.

From dashcams and commercial fleet cameras to drones, event data recorders (EDRs) and vehicle infotainment systems, forensic engineers have access to more information than ever before. The challenge isn't simply collecting the data but validating it, interpreting it correctly, and integrating it with the physical evidence to develop scientifically supported conclusions.

Cameras Are Changing the Landscape

Video evidence has become increasingly common in accident investigations.

According to a recent survey, nearly 30% of U.S. drivers now use dashcams, a figure that continues to grow each year. Commercial fleets have embraced the technology even more rapidly. One industry report found that 84% of company drivers now have dashcams installed in their work vehicles, representing a 29-percentage-point increase over the previous year. Nearly 65% of commercial drivers also reported that their employers require dashcam use.

Those numbers illustrate an important trend for forensic engineers: many serious crashes are captured on video. Additionally, investigators may have access to recordings from multiple sources, including personal dashcams, commercial fleet cameras, traffic cameras, security systems, and even bystanders' smartphones.

While multiple viewpoints can provide valuable information, each recording captures only one perspective. Camera placement, frame rate, image distortion, lighting conditions, and synchronization all affect what a video can and cannot reliably demonstrate.

For that reason, video evidence is never evaluated in isolation. Engineers compare every recording against the physical evidence and established engineering principles before reaching conclusions.

Preserving Crash Scenes Before They Disappear

Unlike digital evidence, physical crash scenes are temporary.

Vehicles are removed, debris is cleared away, and weather or traffic can quickly alter roadway markings. Important evidence may disappear within hours.

Drones and 3D laser scanners have become increasingly valuable tools for documenting crash scenes before those changes occur. 3D laser scanners allow the capture of tens of millions of individual points in 3D space, allowing for millimeter-level 3D environments to be digitally preserved in full detail and color. In conjunction with 3D laser scan data, drones allow the capture of hundreds of overlapping high-resolution photographs of an accident scene in detail within minutes. Using photogrammetry, those images can be converted into an accurate three-dimensional model and scaled map-quality imagery of the roadway, vehicles and surrounding environment.

Rather than relying solely on photographs taken from ground level, investigators can preserve the entire scene from above, creating a permanent digital record that can be analyzed long after the roadway has reopened.

Building a Three-Dimensional Reconstruction

A three-dimensional model provides much more than a visual representation of the crash scene. Once the scene has been accurately documented, forensic engineers can integrate additional evidence into the digital model, including dashcam recordings, surveillance footage and other video sources.

This allows investigators to:

• Measure distances, roadway grades and vehicle positions with precision.
• Analyze vehicle speeds and timing throughout the sequence of events.
• Evaluate driver sight lines, visibility and roadway geometry.
• Compare video evidence against physical measurements.
• Virtually revisit the crash scene long after the physical evidence has been removed.

The result is a comprehensive reconstruction built on measurable data rather than assumptions.

Today's Vehicles Generate More Data Than Ever

Many modern vehicles record far more information than most drivers realize.

Event data recorders (EDRs), also known as your vehicle's "black box," may capture information such as collision severity, vehicle speed, brake application, throttle position, steering inputs, seat belt status and other vehicle operating parameters immediately before and during a collision. In fact, Knott Laboratory can retrieve and analyze EDR information on 57 makes and 700 models of vehicles.

In many cases, investigators may also have access to data stored within infotainment systems, connected devices or telematics platforms. Depending on the vehicle, these systems can provide additional context about navigation activity, communications, connected devices or vehicle operations before the collision.

Each source contributes another piece of the puzzle, but none tells the entire story on its own.

At Knott Laboratory, engineers evaluate digital information alongside vehicle inspections, roadway evidence, scene documentation and witness information to determine whether all available evidence supports a consistent reconstruction.

Technology Supports the Science

As vehicles become more connected and video becomes increasingly common, accident reconstruction continues to evolve. Digital evidence has expanded the tools available to forensic engineers, but it has not changed the scientific principles that guide every investigation.

The role of the forensic engineer is to objectively analyze every available source of evidence, understand the strengths and limitations of each, and determine whether the facts support a particular conclusion.

Whether reconstructing crashes involving passenger vehicles, motorcycles, commercial trucks, pedestrians or bicycles, Knott Laboratory combines decades of forensic engineering experience with today's advanced investigative technologies to provide objective, scientifically supported analyses for a host of clients, from insurance providers to attorneys.

Technology captures the evidence, and science gives the evidence meaning.

With the "silver tsunami" creating significant attrition, it's no secret that the insurance workforce is shifting. Organizations across the insurance ecosystem are struggling to attract the talent needed to fill critical gaps and prepare for what comes next.

Solving for these challenges requires us to think about people differently – not just as applicants for today's openings but as potential contributors to tomorrow's workforce. By shifting how we evaluate talent – from what someone has previously done to what they're capable of becoming – and then by cultivating traits like learning agility and future capability internally, organizations can build a robust and successful workforce for the future.

Identifying and Prioritizing Transferable Skills

The traditional approach to recruiting often looks backward, focusing on titles held, industries worked in, and responsibilities already performed. Those things matter, but they should not be the only measure of a candidate's value. What if instead we look for transferable skills? This does not mean lowering standards. It means expanding the lens. It means asking better questions about capability, judgement, adaptability, and readiness. Have they led a volunteer group? Did they mention a project that required impressive problem solving? Once these skills are identified, it's then up to talent leaders to determine how those skills can be built upon to develop the optimal employee.

Take a mid-career candidate – someone who's very accomplished, has many degrees, and has been a leader, but now wants to move to something different. Some hiring managers wouldn't even consider this individual for certain roles, concerned they're too skilled. But it's a mistake to dismiss a candidate as a poor fit for a position without trying to understand why this person may be interested in the role in the first place. Maybe they want a different pace. Maybe they are looking for a new challenge. Maybe they want to apply their experience in a more focused way. Maybe they're being intentional about the next chapter of their career.

Hiring for potential allows organizations to recognize transferrable skills, avoid overlooking strong talent, and build teams with broader experience, perspective, and capability.

Pinpointing Learning Agility as a Signal of Future Capability

One meaningful way to spot learning agility is to listen for how candidates connect their past experience to future contribution. Candidates who are committed to growth often show it through action: coursework, certifications, self-directed learning, stretch assignments, volunteer leadership, or intentional career pivots. They are usually able to explain not only what they have done, but what they have learned and how they are applying it. Look for candidates who are taking courses, initiating self-training, or working on degrees that have the capacity to expand or re-route their career trajectory. Listen to how they frame the skillsets they have developed in previous roles.

This perspective shift works on the talent development side too. Personally, I always encourage employees to be visible about their skills, interests, and growth goals. Internal profiles, skills-based assessments, and development conversations can help surface capability that may otherwise go unnoticed. This kind of professional curiosity and up-skilling are important but so are previously demonstrated traits like resilience, adaptability, and change readiness – all things to identify and cultivate. In my experience, some of the strongest talent decisions happen when we look beyond the obvious and pay attention to the qualities that signal readiness: curiosity, accountability, sound judgment, resilience, and the ability to lead through change.

In the same way we look for skills when screening resumes and talking with candidates, talent leaders should be thinking about these elements for internal roles and promotions as well. The skills people identify in themselves and how they rate themselves on assessment tools can inform development opportunities and identify organizational fits that might not otherwise have been obvious.

Spotting and Maximizing Leadership Effectiveness Skills

Manager effectiveness remains one of the most persistent and consequential challenges in human resources. If high-performing employees are thrust into supervisory or managerial roles without training on how to meet new expectations, it's not surprising when they don't succeed. By identifying transferable skills already present in the workforce and pairing them with practical tools, coaching, and accountability, organizations can develop leaders who are prepared, not just promoted.

The process of creating courageous leaders is not solely about preparing employees for a leadership role; it's about creating a courageous-leader mindset. A courageous-leader mindset means being willing to ask thoughtful questions, listen before concluding, challenge assumptions, and recognize where bias or blind spots may influence decisions.

Layered on top of this mindset, practical professional development and training programs can help foster new skillsets and build practical capability. Whether through monthly manager development sessions, focused learning labs, or intensive supervisor boot camps, organizations can build practical leadership capability in a way that is structured, consistent, and immediately applicable.

Building Tomorrow's Workforce Starts with Seeing Talent Differently

Not every individual with transferable skills will be the right fit for every role, and thoughtful selection still matters. But when HR leaders prioritize potential, cultivate internal capability, and apply consistent judgment, we create stronger pipelines, more adaptable teams, and a workforce better prepared for the future.

We have built a lot of technology around insurance distribution. We have systems for advertising, quoting, underwriting, CRM, lead management, automation, servicing, and now AI. But there is a part of the experience before most of those systems begin that I think deserves more attention.

Someone has to move from "I need insurance help" to "Here is an organized request that another person can actually understand and act on." Today, we often treat those as though they are the same moment.

A shopper fills out a form or expresses interest, and a record begins moving through the system. But the shopper may still not understand who will receive the information, what happens next, or what kind of contact to expect. The insurance professional receiving the request may still have to figure out why the person is shopping, whether the request fits, what has already been explained, and what information is still missing.

That creates work on both sides, and it can make an already confusing insurance-shopping experience feel even more fragmented. The shopper may have to repeat herself. The professional may have to reconstruct context that should have been clear earlier. And the system can still call the handoff successful because information moved from one place to another.

The Middle Layer

I think the industry needs to pay more attention to this middle layer. Not as another marketing funnel. Not as a way to move people faster. As the point where shopping intent becomes an organized insurance request that both sides can understand.

For the shopper, that means clearer expectations. Who may receive the request? What happens next? Is a licensed insurance professional likely to review it? What kind of contact should the shopper expect? The goal is not zero contact. It is clearer, more controlled contact.

For the insurance professional, it means better context. Why is the person shopping? What has already been provided? What still needs to be clarified? Is the request something this professional can reasonably help with? A request should be organized enough that the receiving professional can make a meaningful decision instead of starting from zero.

That is different from simply collecting more fields. More information does not always create more understanding. A long form can still produce a weak request if the person completing it does not know what will happen next or if the professional receiving it still has to figure out the reason behind the request.

Where Consumer Control Matters

This is also where consumer control matters. Insurance shopping is often repetitive, high-pressure, and difficult for the shopper to control. People worry about spam calls, who gets their information, whether they are being sent everywhere, and whether someone licensed is actually going to help them. Those concerns are not separate from the request process. They are part of it.

A better request layer should preserve enough shopper authorization and context that the next step feels understandable rather than surprising. The shopper should know what she is choosing. The professional should know enough to decide whether to engage. And the handoff should carry forward the information needed to continue the experience instead of restarting it.

AI can help here, but AI should not become the story. It can organize information, identify missing details, summarize what the shopper has already explained, and improve request readiness. Those are useful capabilities. But the technology still has to support a clear human process.

If the shopper does not understand what is happening, or the professional does not have enough useful context, making the handoff faster does not necessarily make the experience better. We may simply automate confusion.

What Marketing Doesn't Measure

I came to this problem after spending years around digital acquisition and customer behavior. In marketing, we get very good at measuring clicks, forms, and conversions. Insurance made me look at those actions differently because there is a person inside every one of them, and the action is often only the beginning of what that person actually needs.

That is the problem I am working on through Ensurance, a U.S. insurance-shopping company beginning with auto insurance. We are building around the idea that a shopper should be able to start one organized request with more clarity and control, while participating licensed professionals receive better context around requests they may choose to engage with.

Behind that, we are developing CATE, our Controlled Access Trust Engine, as the controlled-access and request-intelligence layer. Its role is to help structure requests, improve readiness, support controlled access, and create more useful opportunity context while preserving shopper authorization. CATE is being built, and planned AI capabilities are still under development.

The larger point is not about one company. As insurance becomes more automated, the quality of what enters those systems matters more. The industry does not need another way to turn shoppers into records faster. It needs a clearer way to turn insurance-shopping intent into an organized request that both sides understand.

That is the missing layer I think deserves more attention.

The insurance industry's persistent talent crunch goes far deeper than a hiring gap – it's also becoming a knowledge gap, as the accumulated judgment of a workforce built over decades retires faster than it can be replaced.

By the end of 2026, the U.S. Bureau of Labor Statistics estimates roughly 400,000 insurance professionals will have retired in the last half-decade alone. Today's average insurance employee is in their mid-40s to mid-50s, depending on the line of business. One in four underwriters is already over 50, and less than a quarter of our workforce is under 35.

Turnover has climbed from a historical 8-9% to 12-15% industry-wide. Even with a nearly unprecedented 14 months of workforce reductions, as insurance economist Dr. Robert Hartwig has noted, the sector's 3.3% unemployment rate is still far below the national average of 4.4%.

And that's just the demographic aspect to the story. The bigger issues are what insurers lose when long-tenured experts retire - and what we're doing right now to fill the talent pipeline with future innovators and leaders.

The Industry's Best-Kept Secret Was Its On-Ramp

Ask almost any insurance executive how they got their start, and you'll hear some version of the same story. They largely majored in something other than RMI (risk management/insurance) as undergrads. They came to insurance in an entry-level position – a junior-level claims processing role, a customer service line, or an underwriting support seat.

The pay was good but not great, and the title wasn't impressive. But this was accepted as part of the bargain for landing a job that taught them how the business actually worked in a field that offered tremendous stability and growth opportunities.

They learned how a policy is priced by experiencing the underwriting and pricing process in real-time.

They learned how a claim goes sideways by handling complex claims.

Over years, that ground-level exposure became accumulated knowledge, which in turn became true expertise and leadership.

This story isn't merely an exercise in nostalgia; it's the operating model that built our industry's immense knowledge base.

Insurance knowledge has never been something learned just from a textbook. It's acquired by doing the scut-work long enough to understand why the rules exist. This is the foundation for valuable context that later empowers people to spot the exception, catch the fraud, or make the call a model won't – or can't.

It also happens to be one of the industry's most reliable recruiting advantages.

Nevertheless, insurance has struggled for years with a perception problem. Nearly one-third of the global population is Gen Z (birth years 1995-2012, per Pew Research), yet 79% of respondents in this age cohort say they've never considered working in insurance because it looks boring or overly corporate, according to a 2025 Cake & Arrow survey.

Beyond this, insurance is marketed in such a way that it's either an abstraction (or worse, the bill their parents complain about bitterly). The best we can hope for is that kids who are not old enough to drive are able to embrace and internalize geckos, "Mayhem," and "Bibberty" as past generations learned to recognize Mickey Mouse or Joe Camel in their pre-literate phases.

The insurance career on-ramp – the promise that you can enter the field without an advanced finance degree and build a real career in an essential, thriving industry that's focused on the future – is one of insurance's strongest attractions as a career destination. Dismantling this at precisely the moment we need it most would be a catastrophic mistake.

Where AI Actually Helps, and Where It Quietly Doesn't

This is not an argument against AI. Applied strategically, AI eliminates repetitive, low-value work (data entry, document review, first-pass triage, routine correspondence) that used to take hours or days. Agentic AI is proving useful for executing entire multi-step processes, like claims intake or policy servicing, without a person managing every stage. This efficiency creates a higher standard of service and customer satisfaction.

But there's a vast difference between using AI to remove drudgery and AI inadvertently undermining insurance's "farm system," the roles in which fledgling insurance experts build knowledge and wisdom. Companies eliminating entry-level claims or underwriting support positions to fatten margins as AI scales to handle the volume aren't merely cutting costs. They're turning off the talent pipeline that credentialed underwriters and licensed adjusters flow through.

Industry hiring forecasts make the stakes plain: projections show underwriting and claims roles shrinking even as demand grows for developers, actuaries, InfoSec specialists, and data scientists. But those burgeoning roles don't carry the state licensing and credentialing requirements that underwriting and claims work does – which means the very entry-level positions insurers are cutting are the ones that produce the credentialed underwriters and licensed adjusters that insurance businesses (and in many states the law), require.

This creates more than a talent or experience shortfall: it potentially opens a compliance gap, since carriers need licensed, credentialed professionals to legally price and adjust risk. Underwriters typically spend years earning their Chartered Property Casualty Underwriter (CPCU) or Certified Insurance Counselor (CIC) designations, building the on-the-job experience that shapes how carriers price and structure risk.

Claims adjusters face even greater educational and regulatory hurdles: 34 states require independent adjusters to hold a license, and many additional states require staff (in-house) adjusters to be licensed as well, meaning a person cannot legally handle a claim in a majority of the US without passing state exams and meeting continuing education requirements.

Training alone does not clear that bar. Eliminating entry-level roles in which claims workers accumulate the experience and knowledge needed to earn these official designations doesn't merely remove valuable institutional memory; it endangers insurers' future ability to compete in increasingly crowded markets.

Are We Eating Our Seed Corn?

FINRA's 2026 oversight report isolated AI hallucinations as a specific compliance risk, noting the failure isn't that systems misunderstand questions – it's that they don't recognize when they don't know the answers, replacing probability with, well, "plausibility" – answers that "look" right to the untrained eye.

Some carriers have started layering additional review AI agents into the process specifically to catch these errors before they reach a human operator, and early research shows this step can cut hallucination rates meaningfully. But someone still has to know what a correct answer looks like well enough to catch what even these systems miss.

Today, those people are the ones who came up through the ranks. Tomorrow, absent this entry-level talent pool, the industry might lack adequate numbers of human experts in the loop to verify and correct AI outputs just as our industry leans harder into the technology.

If insurers de-emphasize developing talent internally and try to fill gaps by hiring experienced insurance workers from other companies, they'll be competing in a tightening market in the least cost-effective way possible.

The USBLS projects roughly 21,500 claims job openings per year over the next decade. Experienced underwriters and adjusters don't grow on trees – and the industry is shedding them faster than it's producing them.

A carrier that starves its own farm system hasn't opted out of the problem. It's locked itself into paying retail for talent that competitors are still growing at wholesale, ensuring a future of longer hiring cycles and higher salary floors.

How Insurance Leaders Can Act Today to Safeguard Their Business for the Long Haul

• Protect roles that build judgment while you automate tasks around them. View entry-level positions for what they teach, not just what they cost, and redesign them so AI absorbs the repetitive tasks while people spend more time on work that builds understanding.
• Use AI to raise the ceiling on early-career work, not lower the floor. Give newer employees tools that let them take on higher-value problems sooner, with human and AI oversight built in, rather than tools that perform the work itself for them and leave nothing to learn.
• Keep a trained human expert in the loop on every agentic workflow that matters. Regulators are already moving this direction. More than 20 states have adopted the NAIC Model Bulletin on AI Systems as of mid-2026, and examiners expect documented human oversight on high-risk decisions.
• Treat internal development as a cost strategy, not just a culture initiative. Every role built internally is one you don't have to buy at a premium in tight talent markets.
• Build succession plans ahead of the exit wave, not after it. Know which roles are at risk this year and who could be ready to step up with the right support before those seats become external searches.

AI is helping insurers work faster and more efficiently than ever. Today's leaders are responsible for recognizing when and where we should solidify and expand people's knowledge, rather than reflexively eliminating roles that have traditionally added value across the insurance lifecycle.

Sources

U.S. Bureau of Labor Statistics data, via PropertyCasualty360, "Insurance industry talent shortage is imminent," June 2025

Insurance Thought Leadership, "Insurance Industry Faces Critical Talent Shortage," May 2026

Insight Global, "Retiring Underwriters Are Creating An Insurance Talent Shortage"

Sonant, "Insurance Staffing Shortage 2026: Crisis Data & AI Solutions," April 2026

Sonant, "Insurance Agency Talent Shortage: Solutions for 2026 & Beyond"

PYMNTS, "Insurance Industry May Be Unprepared for Agentic AI Risks," 2026

Notch, "AI Hallucinations in Insurance: Risks & Fixes"

actuary.info, "Adversarial Self-Critique Rewrites AI Underwriting Governance," June 2026

Cake & Arrow, "Why Gen Z Is Ambivalent About Working in Insurance," Oct. 2025

NAIC, "State Licensing Handbook, Chapter 18: Adjusters"; ADbanker, "Insurance Adjuster Licensing Requirements by State," 2025-2026

Kore1, "Insurtech Hiring Trends 2026" (kore1.com/insurtech-hiring-2026) https://www.pewresearch.org/short-reads/2019/01/17/where-millennials-end-and-generation-z-begins/

I don't know about you, but my bingo card for generative AI did not include its wiping out the jobs of Kenyans who had for years been writing essays for cheating American college students. Yet here we are, according to an article in the New York Times last week. 

That story, plus other surprises from the early years of generative AI, should remind us that our ideas about how the AI revolution will play out should be held lightly. We need to be able to drop ideas immediately when they turn out not to be true and pursue new opportunities as soon as they present themselves.

Let's have a look. 

The New York Times article says Kenya had already seen one wave of job losses to AI. Thousands had been employed transcribing recordings of meetings, but such transcriptions became an early task for AI and began to disappear in 2017. Still, ChatGPT created a shock when it debuted in late 2022. At the peak, some 40,000 people in Nairobi were being paid to do homework for others, but almost all those jobs have diminished in scope or disappeared. Why would a student pay someone to write a paper that AI would generate for free?

The article says that "what jobs remain in essay writing are mainly for 'humanizers,' who edit papers written by A.I. to avoid detection from the cheating software used by many universities."

We've seen plenty of other surprises because the backlash against AI caught Big Tech unawares. Any company associated with the hyperscale data centers being used for AI now faces massive pressure because of their water and electricity needs and because lots of people just don't want to live or work anywhere near them. Flock Safety had achieved an $8 billion valuation as it installed cameras around the U.S. to help police track cars used in crimes but has recently stirred up horror about the possibility of a surveillance state.

At a smaller scale, we've seen an Air Canada chatbot go rogue and offer a discounted fare that the airline had to honor. Zillow set up an AI to monitor home prices and purchase undervalued properties, only to find, some $500 million in losses later, that the project was a flop. IBM and McDonald's set up an AI to handle drive-thru orders but canceled the project after customers posted funny, and highly embarrassing, videos of mistakes on orders — such as adding hundreds of dollars in chicken nuggets or bacon toppings on ice cream.

It's easy to imagine where other, important surprises might arise, too. AI will surely have some effect on where people live, because it will change the dynamics of remote work. AI already seems to be having some effect on car sales, because data centers are contributing to increasing electricity prices, which make electric vehicles less enticing. 

In the much-quoted framework from the late U.S. Secretary of Defense Donald Rumsfeld, there are lots of unknowns out there about AI. We can't know much about the unknown unknowns. But we can certainly prepare ourselves for the known unknowns, and I'd argue that all the surprises that have hit us so far — job disappearances, including in Kenya; technology glitches; backlash to AI — fall into the "known unknown" category. (I'd further argue that known unknowns were Rumsfeld's undoing. He should have known that loads of problems could arise after the U.S. occupied Iraq and should have prepared far more for contingencies, rather than assume that Iraqis would place flowers in the barrels of U.S. tanks.)

The history of technology is full of bad ideas — and of pivots toward transformative ones. Alexander Graham Bell initially thought the telephone would be used to broadcast into people's homes — you'd pick up your phone and listen to, perhaps, a symphony orchestra. (Bell also suggested that the standard greeting should be, "Ahoy!" Hmmm....) YouTube began as a video dating site. Slack was a chat channel for a multiplayer game that got left by the wayside. PayPal started as security software for the late, great Palm Pilot, which had a digital payment feature. And so on.

My favorite pivot actually comes from well outside the tech world — from chewing gum. A young man who moved to Chicago in 1891 sold his father's line of soap and offered small amounts of baking soda for free as an incentive to buyers. The baking soda turned out to be more popular than the soap, so he began selling baking soda. This time, he offered some free chewing gum as the incentive. That gum became so popular that, 130 years later, I have a pack of William Wrigley's spearmint gum on my desk.

So my conclusion from the job losses in Kenya (and elsewhere), and from the many other surprises we've seen, is that mistaken ideas about AI are fine — as long we hold them lightly and can move on from them quickly. Think big, because the opportunities with AI are enormous, but learn really fast. 

Cheers,

Paul

Insurance is known as an industry built on caution and calculated decision-making. That isn't going to change, nor should it. But caution becomes a problem when it turns into avoidance.

As AI becomes more embedded across the insurance industry, the greatest risk for many organizations isn't moving too quickly. It is moving too slowly — because some leaders are waiting for the technology to be perfect before they are willing to engage with it at all. That is Passive Sabotage.

Passive Sabotage is the art of dismissing a capability simply because it isn't as good as you think it should be. Does AI code as well as an experienced engineer? Questionable. Does it do it 30 times faster? Yes. Does it know how to tackle new and complex functionality without weeks of meetings, spiking, and exploration? It does. The result is working features in hours or days, not weeks.

Passive Sabotage shows up in small decisions that undermine progress. Abandoning an AI tool after an imperfect result. Delaying a pilot until every problem is solved. Or defaulting to manual processes just because that's what's familiar.

The Real Consequences of Hesitation

This hesitancy comes with real consequences. Organizations that take this approach may miss critical information, move more slowly than competitors, keep experienced employees buried in administrative work, and fail to make full use of the data already available to them. Over time, the gap will widen between companies that are learning how to use AI responsibly and those still debating whether the technology is ready.

Insurance organizations do not need to force AI into every workflow to realize its value. Decisions that carry legal, financial, and human consequences still require consistency, oversight, and experienced judgment. The practical opportunity is to use AI to help teams move faster, reduce manual work, surface information, and test ideas that might otherwise never move beyond discussion.

This distinction matters because much of the skepticism around AI starts with the wrong question. Leaders often ask whether AI can do an entire job as well as a human expert. The better question is whether AI can help a team get to a usable answer, workflow, prototype, or insight faster than they could on their own. In many cases, it can.

For example, AI can summarize claim notes, extract key information from documents, identify missing data, draft communications for review, and help teams prototype new workflow ideas. None of these use cases require handing authority to a fully autonomous system — they require placing intelligence at the right point in an existing process, with the right guardrails, so people are better equipped to make decisions.

Personal Proof

I have tried to demonstrate this through my own work rather than just advocate for it. In the past six months, personally — in between meetings, on planes, nights and weekends — I built:

• A complete marketplace platform with custom integrations, installation workflows, reporting, and analytics. Historically, I had spent over $150,000 annually on white-labeled embeddable marketplaces for our SaaS products. No more. Built, tested, security-reviewed, and deployed in under four weeks.
• A centralized enterprise-wide content and knowledge library with a fully integrated newsfeed to deliver personalized updates across the organization — all AI-enabled summaries and agents — replacing a platform that had cost over $180,000 annually.
• An Outlook add-on built end-to-end in 90 minutes using publicly available API documentation. Another two to three hours of refinement based on testing and feedback.

Were those outputs perfect on the first pass? No. But they were usable — and they were created dramatically faster than they would have been through a traditional process.

Speed changes how organizations learn. Teams can move from idea to prototype in a fraction of the time. They can react to something tangible instead of spending weeks writing requirements for something they have not yet seen. Leaders can test concepts, refine workflows, and make more informed decisions earlier in the process.

This is especially important in insurance, where many organizations are led by experienced executives who did not enter the industry because they were passionate about technology. For many, technology has always been a means to an end. That is understandable, but AI requires leaders to build a new level of familiarity. They do not need to become engineers, but they do need to understand where the technology can create practical value.

Where Organizations Get Stuck

This is where many organizations get stuck. But companies moving ahead are not waiting for AI to become flawless. They are learning where it works, where it needs oversight, and how it can fit into existing processes.

Governance must be part of that process from the beginning. In insurance, requirements can vary by state, business line, client, and use case. Organizations need the ability to decide where AI is used, where it is not, and the option to quickly adapt capabilities if requirements change.

Because of this, modularity is essential. A rigid AI deployment can create risk and limitations, but a configurable approach allows organizations to apply AI where it makes sense, keep human judgment where it is needed, and adapt as the regulatory environment changes. In insurance, where a regulatory change can shift how you apply AI overnight — by state, by client, by risk type — a deployment you can't reconfigure by rule is a liability.

Trust will remain one of the biggest barriers to AI adoption. Many professionals are still getting comfortable with the technology in their everyday work. Technology providers and industry leaders have a shared responsibility to deliver tools that feel useful rather than threatening, and to be honest about where AI needs human oversight and where it does not.

AI does not have to be perfect to be valuable. It does not have to replace expert teams or solve every problem to create a competitive advantage. Insurance leaders who understand that will learn faster and make better use of the information already in front of them.

The companies getting ahead are learning where AI works, where it needs oversight, and how it fits into existing operations — through direct use, not by delegating the evaluation and waiting for a recommendation. If you're a "leader" riding side-car to the opinions of others on this, your days in that role are numbered. It's time to begin using AI yourself.

The insurance industry has been arguing all summer over whether AI liability can be insured. Most carriers have decided - at least for now - that it can't. They are narrowing policy language on the GL, cyber, and product liability forms, adding AI exclusions, trying not to repeat the "silent cyber" problem from last decade. A smaller group of MGAs, Lloyd's coverholders, and reinsurer-backed programs has started writing affirmative AI liability coverage. Their approaches have little in common: performance warranties, adversarial testing, governance reviews, litigation-based models or just AI endorsements bolted onto existing cyber policies.

AI liability claims are still sparse, but the risk exposures are concrete: an autonomous agent that approves $50,000 in payments it shouldn't have, or an AI hiring tool that screens out protected classes. The potential buyers: any company deploying AI where errors hit third parties. Healthcare, finance, legal, HR screening.

These are different perils, different policy forms, different loss dynamics, and that's part of what makes measurement so hard. We're in the early innings of AI insurance. Some skeptics say you can't price it at all. I wouldn't go that far. But what are we actually trying to insure, and on what basis

Who's Writing AI Liability Today

The market is small enough that you can name the entire first cohort.

Munich Re launched the first dedicated AI insurance products, aiSure, as a performance warranty: if the model drifts below defined accuracy thresholds or produces discriminatory outputs, the policy pays. They've been at this since 2018 and remain one of the few reinsurers with a dedicated AI liability product in market.

At Lloyd's, three coverholders have launched AI liability products. Armilla AI offers standalone coverage, with underwriting informed by over 500 AI system evaluations. AIUC takes a different path: it certifies an AI model first, then adds insurance. Their AIUC-1 framework puts systems through thousands of adversarial simulations. The insurance is written on Beazley paper, with ElevenLabs, an AI voice generation company, as their first public customer. Testudo builds its underwriting off AI litigation data, though the dataset is still thin.

Corgi started offering an AI and algorithmic liability endorsement on top of D&O, E&O, and cyber. Cowbell added AI-specific underwriting factors to its cyber risk-rating framework in July. That's probably the near-term path for most cyber MGAs: AI bolted onto cyber, not a standalone line.

And then there's the rest of the market. Technology companies can buy standard Tech E&O from Vouch, Hartford, or Hiscox, with an AI endorsement added. No AI-specific risk assessment, no evaluation of model behavior. Premium is based on revenue, headcount, vertical, prior claims, the same way you'd price a SaaS platform or payroll tool.

A handful of players are experimenting with AI-specific underwriting. A much larger market isn't measuring AI risks at all. Total dedicated premium for AI liability: immaterial.

From Signal to Pricing

AI liability is one of the hardest emerging risks to insure.

Parametric insurance works because the trigger is a pre-agreed, observable, independently verifiable number: a NOAA weather station, or a cat model from Moody's RMS.

AI evaluation scores don't work that way. A hallucination rate changes with the test set, the evaluator, the model version, and the business context. Armilla and AIUC evaluate AI models, but when the evaluator is also the insurer, the data isn't independent anymore, by definition.

What underwriters need is a validated chain: an observable AI signal that correlates with loss frequency, that maps to expected severity, that can be modeled across a portfolio. I haven't seen a public demonstration of this chain from signal to pricing.

Gallagher Re laid this out in their June 2026 report "Anthropic's Fourth Way": current AI evaluation methods "were not designed for underwriting and are not fit for that purpose." Benchmarks measure how models perform on controlled tests. But losses happen in deployment, not in testing. Their conclusion:

"If a model cannot be tested, insurers end up pricing uncertainty rather than risk."

Without better evaluation, that leads to two failure paths: AI losses absorbed silently into existing policies until carriers exclude them, or standalone products launched without foundations that collapse after early losses.

Without independent model evaluation, the market defaults to underwriting governance: deployment approval processes, human oversight. Most AI underwriting is done that way today, but I'm not convinced that is enough. Cyber insurance tried governance-based underwriting for years, remember? Applicants checked "yes" on the MFA question, and half the time they didn't have it deployed properly. It took a catalyst, the ransomware wave of 2020/21, to force the industry to verify independently what applicants were telling them.

Where the Cyber Comparison Breaks Down

BitSight and SecurityScorecard built outside-in security scores for cyber, based on a client's open ports, SSL certificates, and malware infections. The same approach doesn't work for AI risk. A publicly visible chatbot might reveal something about prompt-injection resilience. It reveals nothing about the agent's authority, data flows, approval thresholds, or what happens when it makes a wrong decision. No external attack surface to scan.

There is a trust problem underneath all of this. Policyholders resist giving insurers access to internal data. They worry it might get used against them in a coverage dispute. And insurers have their own reasons not to look. I learned this at a cyber MGA: if you discover a vulnerability in a client's network through an internal vulnerability scan and don't act on it, you're exposed to E&O claims. Better not to know. For AI, it gets worse, as the vulnerabilities are harder to define, and there's no patch to deploy.

The Missing Layer

What the nascent AI insurance market needs is an independent measurement layer, something that takes technical AI evidence and turns it into data an underwriter can price from.

AI governance platforms like Holistic AI and Credo AI score AI systems on bias and robustness. They were built for compliance teams, not underwriters, and tell you whether a system meets a regulatory standard. They don't tell you the likelihood and cost of a liability loss.

Neither side can build this layer alone. The insured won't share data they fear could be used against them. The insurer faces the E&O problem I described. This layer needs to sit between them, the way a credit rating agency sits between borrower and lender.

AI insurance will need at least one credible, independent source of underwriting-grade evidence. Whoever builds that becomes infrastructure.

What would you trust as underwriting evidence for an AI agent: pre-deployment testing, runtime telemetry, an independent rating, a contractual warranty, or some combination?