AI Insights Mar 20 2026

Mar 20, 2026

insuranceindustry.ai article hero image

AI Insights: March 20, 2026

Your weekly analysis of AI developments in insurance.

WTW Survey: Insurers Using Advanced Analytics Are Outperforming. The Gap Is Widening.

WTW released its 2026 Advanced Analytics and AI Survey this week, and the headline finding should concentrate the attention of every insurance executive who has been waiting for proof that AI investments actually pay off: P&C insurers in North America that invested more heavily in advanced analytics and AI achieved combined ratios six percentage points lower and premium growth three percentage points higher than slower adopters between 2022 and 2024.

Those are not marginal differences. A six-point combined ratio advantage over a three-year period represents a structural competitive gap that compounds over time. The carriers at the front of this curve are not just experimenting. They are pulling measurably ahead in the metrics that matter most to boards and investors.

The survey of 59 P&C insurers found that predictive rating models have become essentially universal, with close to 80% of carriers already using advanced rating and pricing models and another 11% planning near-term implementation. The claims function, historically slower to adopt, is signaling aggressive expansion. Although only about a third of carriers currently use advanced analytics for fraud detection or severity assessment, those figures are expected to reach 65-70% within two years. Straight-through processing in claims workflow automation is planned by 36% of respondents, a significant jump from the current 14%.

On generative AI specifically, more than half of respondents report already using large language models, with another 29% planning adoption within two years. The adoption curve for LLMs in insurance is steepening faster than many industry observers expected even 12 months ago.

Why This Matters:

This is the first major industry survey to attach hard performance numbers to the analytics adoption gap. Prior surveys measured intent and experimentation. WTW measured outcomes. The six-point combined ratio advantage and three-point premium growth differential give executives something they have been asking for since the AI conversation began: evidence that investment translates to measurable competitive advantage.

The data also validates what WTW’s Laura Doddington described as the shift from competitive advantage to essential requirement. When nearly 80% of carriers are already running predictive rating models, the question is no longer whether to adopt. It is whether you can afford to be in the remaining 20%.

Strategic Implications:

For carriers still in early-stage analytics deployment, the WTW data reframes the decision calculus. The risk of investment is now measurably smaller than the risk of inaction. A six-point combined ratio gap is the kind of structural disadvantage that erodes market position year over year and becomes progressively harder to close.

For agencies and brokers, these findings have a secondary implication worth noting. Carriers with more sophisticated analytics are going to become more selective about distribution partners. The carriers pulling ahead on combined ratio and premium growth will increasingly expect their agency partners to deliver cleaner data, faster submission turnaround, and higher-quality risk selection. Agencies that cannot contribute to that value chain will find themselves competing for the carriers who are falling behind.

Anthropic’s Labor Market Study Introduces a New Way to Measure AI Displacement Risk. Insurance Should Pay Attention.

Anthropic published a research paper this month introducing a new measure of AI displacement risk called “observed exposure,” and the methodology and early findings are more relevant to insurance workforce planning than the typical academic exercise.

The approach combines theoretical AI capability scores with real-world usage data from Anthropic’s Claude platform, weighting automated and work-related use cases more heavily than augmentative or casual use. The key insight: AI is far from reaching its theoretical capability. Actual task coverage remains a fraction of what is technically feasible. But the gap is closing, and it is closing unevenly across occupations.

The most exposed occupations include computer programmers (75% task coverage), customer service representatives, and data entry workers. At the other end, 30% of the workforce has zero coverage because their tasks are too infrequent in AI usage data. The study found that occupations with higher observed exposure are projected by the Bureau of Labor Statistics to grow less through 2034. Workers in the most exposed professions tend to be older, female, more educated, and higher-paid.

On the critical question of whether AI has already affected employment, the findings are nuanced. The researchers found no systematic increase in unemployment for highly exposed workers since late 2022. However, they found suggestive evidence that hiring of younger workers (ages 22-25) has slowed in exposed occupations, with a 14% drop in job finding rates compared to 2022. This echoes findings from a separate study by Brynjolfsson et al. using ADP payroll data.

Why This Matters:

The “observed exposure” framework offers insurance executives a more grounded way to think about workforce impact than the breathless predictions that dominate headlines. The distinction between theoretical capability and actual deployment is critical. Many tasks that AI could theoretically perform at twice the speed are not being automated because of regulatory constraints, software integration requirements, or verification steps. That gap is where insurance lives. The industry’s regulatory overlay, fiduciary obligations, and data governance requirements create friction that slows the automation curve relative to less regulated sectors.

The finding about young workers is particularly relevant for an industry already facing a well-documented talent pipeline problem. If AI-exposed occupations are hiring fewer entry-level workers, the insurance industry’s ability to replenish its aging workforce through traditional pathways may be further constrained. The pipeline is not just shrinking because young people are not choosing insurance. It may also be shrinking because the entry-level positions that traditionally served as on-ramps are being absorbed before new workers arrive.

Strategic Implications:

For carriers and large agencies evaluating workforce strategy, Anthropic’s framework suggests a more targeted approach than either “AI replaces everyone” or “nothing changes.” The most productive planning starts by mapping which tasks within specific roles have high observed exposure and which require the kind of judgment, regulatory knowledge, and relationship management that AI does not currently cover. The roles most likely to change are not the ones that disappear. They are the ones that shed their routine tasks and become concentrated around judgment and oversight.

For HR leaders specifically, the young-worker hiring data warrants attention. If entry-level positions in customer service, data processing, and administrative support are being absorbed by AI tools, the traditional career ladder that moved new hires from transactional roles into more complex work may need to be rebuilt around different entry points.

Jensen Huang Envisions 100 AI Agents per Employee. The Insurance Implications Are More Immediate Than You Think.

At Nvidia’s GTC conference this week, CEO Jensen Huang offered a vision of Nvidia’s future workforce that should register across the insurance industry: 75,000 human employees working alongside 7.5 million AI agents. That is a 100-to-1 ratio of agents to humans, operating around the clock on tasks that do not require human judgment.

Huang also unveiled Nvidia’s open Agent Toolkit, a development platform designed to help enterprises build and deploy their own AI agent fleets. Companies including Adobe, Palantir, and Cisco are already working with the toolkit. The announcement follows a McKinsey finding from November 2025 that 62% of organizations were at least experimenting with AI agents, though nearly two-thirds had not yet begun scaling.

Huang framed agents not as replacements for human workers but as force multipliers handling routine execution while humans focus on higher-order decisions. McKinsey’s own CEO, Bob Sternfels, disclosed that the firm now has approximately 25,000 AI agents working alongside its 40,000 human employees.

Why This Matters:

The agent conversation has shifted from theoretical to operational faster than most industries anticipated. Agentic AI differs from the chatbot and generative AI tools that dominate current insurance discussions. Where a chatbot responds to prompts, an agent autonomously executes multi-step workflows: researching prospects, extracting data from applications, comparing contracts against policies, reconciling commission statements. These are the exact use cases the ACT Tech Trends report identified for independent agencies just two weeks ago.

The 100-to-1 ratio is a forward projection, not a current reality. But the direction matters more than the timeline. If the companies building AI infrastructure are designing their own operations around agent fleets at that scale, the technology platforms the insurance industry relies on will be built with that architecture in mind. The question for insurance is not whether agents are coming. It is whether carriers, agencies, and TPAs are building the data foundations and workflow documentation required to deploy them effectively when they arrive.

Strategic Implications:

For carriers, the Nvidia announcement accelerates the timeline on agentic AI readiness. The carriers that have invested in clean data architectures, documented workflows, and API-connected systems will be the first to deploy agent-based automation at scale. The carriers still running on fragmented legacy systems and manual processes will face a compounding disadvantage as agent toolkits become commodity infrastructure.

For agencies, the agent paradigm reinforces a message this newsletter has returned to repeatedly: the agencies that have documented their processes, standardized their data handling, and invested in workflow clarity are building the foundation that agentic AI requires. An agent cannot automate what has not been defined. The agencies treating process documentation as overhead are actually deferring the prerequisite for their own modernization.

Block’s 40% Workforce Cut Sparks a Bigger Question: Is This the Bellwether?

When Block CEO Jack Dorsey announced in late February that the company was cutting 4,000 employees, roughly 40% of its workforce, and attributed the decision to AI efficiency gains, it triggered the most intense debate about AI-driven job displacement since ChatGPT launched. This week, that debate intensified. Oaktree Capital co-chairman Howard Marks told Bloomberg that most people are underestimating AI’s impact on employment, citing the Block cuts as evidence that displacement is happening in sudden, large-scale events rather than gradual attrition.

Marks was blunt in his assessment, noting that 40% of a 10,000-person company disappeared in a single day because AI could perform the work more cheaply and faster. He suggested that if Block’s move proves successful, other CEOs will replicate the pattern, potentially accelerating job displacement across industries.

The “bellwether or anomaly” debate landed at the center of a broader conversation this week. Meta is reportedly planning its own significant workforce reduction following similar logic. A Darden Business School analysis noted that Block’s cuts may also reflect COVID-era overhiring rather than pure AI displacement. And a Harvard Business Review study found that AI mandates at some companies are actually intensifying workloads rather than reducing them, creating what researchers call “AI burnout.”

Why This Matters:

For insurance executives, the Block episode crystallizes a question the industry will need to answer in the next 12 to 24 months: when AI tools demonstrably improve productivity in specific functions, what is the appropriate workforce response? The insurance industry’s version of this question is already playing out in the Aon/Jacobson labor data covered in last week’s newsletter, where the share of insurers holding headcount steady hit a 15-year high while revenue expectations remain strong.

Block’s experience also illustrates the credibility risk of overstating AI’s role. Bloomberg reported that critics accused Dorsey of “AI washing,” using the AI narrative to justify cuts that were actually about cleaning up pandemic-era bloat. That distinction matters for insurance leaders considering similar announcements. Employees, regulators, and investors can tell the difference between genuine productivity-driven restructuring and cost cutting dressed up in AI language.

Strategic Implications:

Insurance carriers considering workforce restructuring around AI capabilities should take two lessons from the Block experience. First, the market rewards clarity and decisiveness. Block’s stock rose more than 20% on the announcement. Second, the market also scrutinizes the underlying logic. If the AI efficiency claims do not hold up operationally, the credibility damage will be significant. The carriers that approach workforce transitions with documented productivity data, transparent communication, and genuine reinvestment in AI-augmented roles will navigate this transition more effectively than those that use AI as rhetorical cover.

MIT Researchers Build AI Model That Predicts Heart Failure Progression a Year in Advance

Researchers at MIT, Mass General Brigham, and Harvard Medical School published results this month from PULSE-HF, a deep learning model that predicts whether a heart failure patient’s condition will worsen within a year, using only standard electrocardiogram data. The model achieved accuracy scores (AUROC) ranging from 0.87 to 0.91 across three independent patient cohorts, which is exceptionally strong performance for a clinical prediction tool.

What distinguishes PULSE-HF from existing approaches is that it forecasts future decline rather than simply detecting current conditions. The model takes an ECG reading and outputs a probability that the patient’s left ventricular ejection fraction will fall below 40% within 12 months, which represents the most severe heart failure classification. Notably, a single-lead ECG version of the model performed as well as the standard 12-lead version, meaning it could potentially be deployed in settings with minimal equipment.

Why This Matters:

Predictive models that forecast health deterioration from routine diagnostic data have direct implications for life and health insurance. If a standard ECG can identify which heart failure patients are likely to worsen within a year, that changes the actuarial calculus around risk stratification, care management investment, and claims reserve adequacy for heart failure-related populations. Heart failure affects approximately 6.7 million Americans and is one of the most expensive chronic conditions for both health insurers and Medicare.

The single-lead capability is particularly significant. It lowers the deployment threshold to primary care offices, rural clinics, and potentially even wearable devices, creating the possibility of continuous or periodic risk monitoring at population scale. For health insurers and Medicare Advantage plans investing in care management and early intervention programs, a validated predictor of near-term decline provides a tool for targeting resources where they are most likely to prevent costly hospitalizations.

Strategic Implications:

For health and life insurers, the PULSE-HF research represents the kind of AI application that moves beyond operational efficiency into underwriting and clinical decision support. Insurers that partner with health systems to integrate predictive tools like PULSE-HF into care management workflows could reduce heart failure-related claims costs while improving patient outcomes. The research also raises longer-term questions about how predictive health data from routine diagnostics will factor into underwriting models, and the regulatory and ethical frameworks that will govern its use.

The Bottom Line

This week’s stories converge on a single theme: the gap between AI’s demonstrated value and the operational readiness required to capture it is becoming measurable, and the penalty for inaction is becoming quantifiable.

WTW showed that P&C carriers investing in analytics are outperforming slower adopters by six points on combined ratio and three points on premium growth. That is not a theoretical projection. It is a measured outcome over a three-year period. Anthropic’s labor market research demonstrated that while AI has not yet caused widespread displacement, the coverage gap between what AI can theoretically do and what is actually being deployed is closing, with early signals that entry-level hiring is already slowing in exposed occupations. Jensen Huang’s 100-to-1 agent vision at GTC described where enterprise infrastructure is heading, while the AutoRek and Insurance Business data confirmed that most insurers have not yet built the data foundations required to get there.

Block’s workforce reduction and the surrounding debate about whether it represents genuine AI transformation or rhetorical cover put the credibility question front and center. And MIT’s heart failure prediction model illustrated how AI is beginning to produce tools that do not just improve efficiency but change the fundamental information available for risk assessment and clinical decision-making.

The connecting thread is accountability. The WTW data makes it harder to defer analytics investment by claiming the ROI is unproven. Anthropic’s research makes it harder to dismiss workforce impact by pointing to aggregate unemployment data. And the agent infrastructure being built by Nvidia, Microsoft, and others makes it harder to treat agentic AI as a distant possibility rather than a near-term planning requirement.

The executives reading this newsletter who are already acting on these signals are building advantages that will compound. The ones waiting for the picture to become clearer should recognize that this week, across multiple authoritative sources, it did.

AI Insights appears every Friday, analyzing AI developments through an insurance lens. For deeper analysis of strategic implications, visit InsuranceIndustry.ai.

By James W. Moore | InsuranceIndustry.AI

Sources:

 

Subscribe to AI Insights Newsletter

AI Disclaimer: This blog post was created with assistance from artificial intelligence technology. While the content is based on factual information from the source material, readers should verify all details, pricing, and features directly with the respective AI tool providers before making business decisions. AI-generated content may not reflect the most current information, and individual results may vary. Always conduct your own research and due diligence before relying on information contained on this site. 

Related Posts

468