U.S. premiums and deposits across insurance categories

The three reporting groups below total approximately $3.98 trillion in reported direct premiums and deposits for 2025. They are shown separately because they cover different risks and use different reporting conventions: property-and-casualty and health figures measure direct premiums, while the life and accident-and-health figure also includes deposits.

The combined amount is premiums plus deposits, not premiums alone. The net P&C figure reflects reinsurance adjustments and is not added again.

Results also differ. P&C reported a 92.9 percent combined ratio in 2025, meaning underwriting costs were below premium income on that measure. Health-statement insurers reported a net profit margin of just 0.4 percent. Neither figure describes every insurer or product. Improving affordability therefore requires attention to loss costs and financial sustainability as well as fees and administration.

Coverage gaps extend beyond home insurance

These measures cannot be added into a single uninsured percentage: populations overlap, years differ, and being uninsured is different from being underinsured. Together, they show demand that clearer advice, simpler enrollment and more economical service could help meet.

Capital, contracts and public responsibility

Customers need reliable payment and fair treatment. Insurers need viable rates and enough capital to absorb volatility. Policymakers need broad access without weakening solvency, while governments must manage public exposure to disasters and other difficult risks. Those interests can conflict: suppressing premiums below sustainable levels can restrict supply, while unrestricted pricing and opaque exclusions can leave essential protection out of reach.

States supervise insurers, licenses and market conduct. Risk-based capital and insurer-funded guaranty systems serve different purposes from federal flood, crop and terrorism programs. Public support is not a blanket taxpayer guarantee for the industry. Agentic systems can make decisions and program coordination easier to inspect, but elected officials and regulators still decide how risk and cost should be shared.

Sustainable premiums and shared value

A premium must fund claims, service, distribution, reinsurance, taxes and capital, with room for a sustainable return. These are not all costs that AI can remove. The opportunity lies in avoidable work: repeated applications, disconnected reviews, billing errors and unresolved handoffs. Agents can reduce that burden across existing policies and new products, making smaller customers economical to serve. Better prevention can also reduce losses.

Long-term cover still offers stability; shorter terms are not inherently cheaper. AI cannot remove medical inflation, catastrophe exposure or uncertain liability. Savings depend on lower costs, better risk management and applicable pricing rules—not simply dividing a policy into smaller pieces or transferring more risk to the customer.

The insurance lifecycle through a ReAct loop

ReAct alternates reasoning, action and observation: examine the case, use a tool, check the result and revise the plan. Unlike a fixed script, an agent can select the next permitted step when evidence is missing or contradictory. It resembles a professional working through incomplete information, without implying identical thinking.

Insurance event → Gather insurance context → Interpret the contract → Reason and plan the response → Coordinate evidence and tools → Verify details → HITL approval where required → Execute the approved action → Feedback loop

An event may be a purchase, changed usage or a loss. The agent gathers context, resolves gaps and checks the result of each tool call before continuing. Feedback updates the case and informs tested improvements; it does not silently change policy rules. Predictable calculations remain in validated rating and transaction systems.

If rented equipment is damaged during a project, an agent checks the applicable policy and rental agreement, requests a missing inspection record and routes a coverage question to an adjuster. After approval, it verifies payment or repair progress. Success is a resolved case, not simply a completed summary.

HITL means human-in-the-loop review before a consequential decision, with evidence, uncertainty and proposed action visible to a qualified person. Routine permitted tasks can proceed within established limits; disputed cover, clinical decisions and consequential exceptions require professional review.

A health-insurance example

Consider a $20,000 knee-replacement procedure requiring hospital admission. An agent can verify benefits, assemble clinical records, identify missing information and track the authorization request. It reduces the administrative work for the patient, hospital and insurer without deciding that an expensive procedure should be rejected.

A prompt telling the agent not to deny solely because of cost is only an instruction. A claims or authorization service that technically blocks an automated denial and requires authenticated clinical review is a guardrail. The reviewer applies the relevant coverage and medical-necessity criteria, records the decision and preserves the patient’s appeal route. The dollar amount here is illustrative, not a quoted treatment price.

Journey one: a family choosing health insurance

A family changing jobs must compare premiums, deductibles, medicines and provider networks while avoiding a gap before new employer coverage starts. The cheapest premium may not mean the lowest total cost. Flexibility here means finding suitable coverage for the transition, not switching medical insurance on only when someone becomes ill.

An agent can help the family:

1. assemble household information, employer coverage dates and enrollment eligibility with permission;
2. check doctors and regular medicines against current plan records;
3. compare eligible plans, subsidies and likely care costs over the transition;
4. prepare enrollment, identify missing evidence and route advice to a qualified person; and
5. confirm start and end dates, preserving continuity and checking benefit or deductible changes.

The family gets a clearer choice; the insurer receives a more accurate application. After enrollment, the same service can gather evidence for a treatment authorization and track qualified review. Health information supports care needs, not unlawful repricing.

Journey two: buying car insurance for the first time

A first-time buyer who drives occasionally may still receive expensive fixed-premium quotes. Before leaving the dealership, they must understand state minimums, lender requirements and optional protection. A mileage-based option could fit better, but a cheaper quote may simply remove cover or increase the deductible.

An agent can:

1. collect driver, vehicle, usage and financing details once;
2. explain required and optional cover in plain language;
3. compare fixed and mileage-based carrier quotes using the same limits and deductibles;
4. estimate bills at low and high mileage, including base charges and eligible discounts; and
5. obtain required approval, confirm proof of cover and review suitability when driving changes.

The buyer sees when fixed or mileage-based cover fits better; the insurer gains reliable usage information. After an accident, the agent can assemble the claim and follow approved repairs, using the policy in force at the time.

Journey three: commercial and cyber insurance for a small business

A small firm needs ongoing liability, workforce and cyber protection, but equipment and project exposures may change month by month. Annual commitments, estimated payroll and repeated paperwork can strain cash flow. Clients still demand certificates and limits, while cyber incidents or liability claims may surface after a project ends.

An agent can:

1. assemble existing policies, contracts, payroll, assets and loss records;
2. distinguish ongoing obligations from temporary exposures and identify gaps or duplicate cover;
3. gather cyber-control evidence and compare annual packages with approved project-specific options;
4. prepare broker submissions and reconcile eligible payroll-linked payments; and
5. confirm approved changes, issue authorized certificates and track expiries and continuing obligations.

The owner matches cover to actual work; insurers gain better exposure evidence. After a cyber incident, the agent can coordinate notices and records for response specialists and claims reviewers. Payroll-linked payments help cash flow but do not eliminate audits or guarantee lower annual premiums.

Four routes into the market

These are alternative business models, not stages every startup must pass through.

1. Workflow supplier: earn software revenue by resolving billing or claims work more effectively than existing tools. The insurer retains product and decision authority.
2. Licensed service business: compete for customers through agent-assisted advice, policy management and claims support. Lower servicing costs can make small accounts attractive.
3. Delegated program: develop a focused offer, such as project-equipment cover, using agents to gather evidence and administer terms within carrier authority.
4. Independent insurer: obtain the licenses, capital and reinsurance to carry risk and compete directly on products, pricing and service.

The regulatory moat is real: agents can reduce compliance work, not replace licenses or capital. Risk-bearing insurers face financial and operational review. Texas P&C insurers, for example, need at least $5 million in capital and surplus, separate from setup expenses. The Uniform Certificate of Authority Application process targets 90 days from a complete application; preparation and follow-up can extend the timeline. These are not an industry-wide startup cost or launch guarantee.

Incumbents can redesign their own services, buy missing capabilities, invest or acquire. Partnerships can combine regulated capacity with technical speed, while competitors pursue customers left dissatisfied. A startup may compete with a broker while using an established carrier’s products. Contracts should make customer ownership, data rights and claims responsibility clear; partnership is a choice, not every startup’s destination.

Augment jobs and make room for more valuable work

Agentic insurance should be built for job augmentation, not job replacement. Handling repeated forms, evidence requests and status updates frees people to advise customers, resolve difficult cases and develop neglected services. The goal is better-quality work and more room for creativity, not simply higher workloads.

New products, prevention and service for smaller customers can generate revenue that supports hiring. Employment growth depends on viable demand and reinvestment, not automation alone. The ILO’s research points toward transformation rather than redundancy for most exposed jobs; how employers redesign work matters.

Turn saved time into career and business growth

1. Redesign work with employees. Allocate saved time to complex cases, customer relationships and neglected projects. Measure useful work and job quality.
2. Build stronger entry-level careers. Continue junior hiring, mentoring and supervised cases. Faster access to context should accelerate professional learning.
3. Share governance. Frontline teams, licensed specialists and compliance should shape workflows, with authority to question, correct and stop an agent.
4. Reinvest in new services and roles. Extend advice, prevention and product development to underserved accounts; hire as sustainable demand grows.

Value must show up in insurance economics

The investment test is useful protection and measurable value after full program cost, not the number of documents processed.

My earlier global economic-value article models an annual insurance opportunity of $150B enabled by 2031, with $100B captured before costs and $80B of net economic benefit. The arithmetic remains $40B from operations and employee capacity + $30B from prevention and new products + $30B from loss, reserve and portfolio improvement − $20B in full program costs = $80B. This is a global author-modeled scenario, not an observed result or a percentage of the $980B U.S. P&C premium figure.

For one insurer, use a workflow-level test:

Net economic benefit from Agentic AI = redeployed capacity value + incremental contribution margin + verified reduction in insured losses − full program cost

Count redeployed time only when it produces useful work; do not count the same gain again in margin or loss savings. Customer savings are not extra insurer revenue. Lower claim payments create value through genuine prevention or reduced leakage, not underpayment of valid claims. Full cost includes technology, integration, controls, employee transition, oversight and failure response.

Track cost per resolved case, customer effort, coverage gaps, newly served accounts and insurer margin. Compare premiums alongside deductibles and uninsured exposure: smaller installments or reduced protection are not automatically savings. Lower operating costs do not automatically become lower premiums either; leadership should decide how verified gains support customer value, employee development, reinvestment and sustainable returns.

The job augmentation dividend

The dividend starts when employees use freed capacity to deliver better advice, resolve claims and reach smaller customers. Customers gain less effort and more dependable protection; insurers gain retention, useful risk evidence and room for profitable growth. Measure skills, career progression and net employment created, not just hours automated.

The wider benefit comes from fewer preventable losses, faster recovery and businesses able to keep investing and employing people. Governments can gain more efficient oversight and less pressure from uninsured losses. But shifting a bill from public budgets to households is not itself a net economic gain. Real resource savings and additional productive activity—not transfers between parties—make the stronger national case.

Turn regulatory variation into shared control infrastructure

Build around versioned, legally reviewed rules. Agents retrieve applicable requirements, gather evidence and run approved checks before quoting, changing cover or handling claims. Conflicts and rule changes go to qualified reviewers before release. The workflow stays reusable; forms, permissions and deadlines change with the case.

U.S. insurance regulation is principally state-led under McCarran–Ferguson, with product-specific federal requirements. NAIC models help coordinate standards but are not automatically binding nationwide; the service must check their actual state implementation and applicable federal rules.

Identify the insurance product and proposed action → Determine applicable states and federal rules → Verify the legal entity and license or delegation → Retrieve effective policy forms and reviewed rules → Check data purpose, pricing and coverage permissions → Obtain required consent and HITL approval → Execute within authority → Record evidence and monitor changes

For example, a mileage-based auto product entering another state can reuse the workflow, but its rating inputs, forms, notices and licensing checks must be reassessed. Approval for one component does not authorize every combination or market. Agents prepare the evidence and track progress; regulators and authorized professionals retain their respective approval roles.

The same records can help regulators inspect decisions, investigate complaints and identify recurring failures. Better evidence reduces repeated preparation for insurers and supports more efficient oversight without transferring approval authority to an agent.

Industry economics, products and markets

• NAIC, 2025 Property & Casualty and Title Insurance Industries Analysis
• NAIC, 2025 Life Industry Commentary
• NAIC, 2025 Health Industry Commentary
• NAIC, Insurance Industry Snapshots and Analysis Reports
• OECD, Gross Insurance Premiums
• NAIC, Telematics and Usage-Based Auto Insurance
• NAIC, Travel Insurance
• The Hartford, Pay-as-You-Go Workers’ Compensation
• Thimble, Business Coverage by the Job, Month or Year
• Guardian, Multiple Life Policies and Laddering Coverage
• Washington Office of the Insurance Commissioner, How Home Insurance Works
• HealthCare.gov, Getting Coverage Outside Open Enrollment
• HealthCare.gov, How Health Premiums Are Set
• Washington Office of the Insurance Commissioner, ACA and Non-ACA Health Coverage
• NAIC, Cyber Insurance Report: Coverage and Claims-Made Terms

Coverage gaps, affordability and public programs

• U.S. Census Bureau, Health Insurance Coverage in the United States: 2024
• Insurance Research Council, Uninsured and Underinsured Motorists
• LIMRA and LOMA, 2026 Life-Insurance Coverage Gap
• Hiscox, 2025 Underinsurance in Small Business Report
• Federal Reserve, Economic Well-Being of U.S. Households in 2025: Housing
• GAO, Homeowners Insurance: Premiums Generally Tracked Inflation but Rose More in Disaster-Prone Areas
• U.S. Treasury, Federal Insurance Office Homeowners Insurance Analysis
• FEMA, National Flood Insurance Program Financial Report
• USDA Risk Management Agency, Program Budget and Financial Highlights
• U.S. Treasury, Terrorism Risk Insurance Program

Regulation, AI and consumer protection

• NAIC, McCarran–Ferguson Act
• NAIC, SERFF: Shared Rate and Form Filing Infrastructure with State-Specific Requirements
• NAIC, Risk-Based Capital
• NAIC, Uniform Certificate of Authority Application: Requirements and 90-Day Review Target
• Texas Department of Insurance, How to Incorporate a Texas Insurance Company: Minimum Capital and Surplus
• NAIC, Guaranty Associations and Funds
• NAIC, AI Model Bulletin Implementation Map
• NAIC, Artificial Intelligence: Insurance Uses, Risks and Governance
• New York DFS, Circular Letter 7 on AI and External Consumer Data
• U.S. HHS, Summary of the HIPAA Privacy Rule
• CMS, Interoperability and Prior Authorization Final Rule
• U.S. Department of Labor, Health Plans and ERISA
• NIST, Generative AI Profile

Agentic systems, economic value and work

• Yao and colleagues, ReAct: Synergizing Reasoning and Acting in Language Models
• Indus Agent AI, Agentic AI: $16Tn in Global Economic Value Enabled and $8Tn Captured
• BLS, Insurance Carriers and Related Activities
• BLS, Occupational Projections and Characteristics
• ILO, Generative AI and Jobs: A 2025 Update

Startup and technology platforms

• WithCoverage
• FurtherAI
• Salesforce
• Sixfold
• Federato
• Bevaya, formerly Roots Automation
• Sprout.ai
• ZestyAI
• Tractable
• Supercede
• ServiceNow
• Guidewire, Insurance AI
• Duck Creek, Agentic Platform