Insurance Benchmarks and the Future of AI Native Professional Support

Executive findings

1. Insurance has resources to invest, but its economics vary sharply by segment. The NAIC reports a 92.9% U.S. P&C combined ratio for 2025. A separate composite of established medical professional liability specialists continued to report a combined ratio above 100%. Aggregate strength can coexist with pressure in professional liability. [1, 2]
2. Licensing defense is a measurable service-cost category. The six selected profession-specific sources report average paid defense expenses ranging from $4,847 to $7,650. These are observations from different insured populations and periods, not a national fee schedule or a ranking of professional risk. [3–8]
3. Insurer AI adoption and AI-native service delivery are different propositions. Existing adoption surveys cover a broad range of AI and machine learning. They do not measure how many insurers have redesigned professional-support workflows around generative AI. [12, 13]
4. Productivity improvements are credible; their transfer into savings must be demonstrated. Research outside insurance shows both benefits and performance deterioration depending on the task. Insurer procurement should measure reviewed output, total effort, and quality. [14, 15]
5. The commercial model matters as much as the model technology. Time saved can become lower invoices, additional capacity, better service, or higher supplier margins. Those outcomes are not interchangeable.
6. AI can strengthen the professional as well as the insurer. Better-informed users may discover benefits, seek earlier assistance, or challenge coverage decisions. Some of this can increase insurer expenditures while improving access and fair treatment.
7. Early partnerships should be narrow and measurable. The most informative starting point is a defined preparation or support service, with clear escalation to counsel and an evaluation that includes users who abandon the workflow.

1 Scope and benchmark methodology

This is a desk-research benchmark report. Sources were selected for direct relevance and traceability, prioritizing regulators, original insurer and program reports, company disclosures, and primary research. No proprietary claims files, patient records, policyholder records, or BoardWise outcome dataset were used. The evidence is current to the research cutoff, but individual datasets often describe earlier years.

The report uses four evidence labels:

All dollar amounts are nominal U.S. dollars unless otherwise stated. Historical comparisons are not inflation-adjusted. Report years must not be confused with the years during which claims closed. A five-year closed-claim dataset does not describe a single year's frequency, and a six-year dataset contains more observation time than a five-year dataset.

The profession-specific observations are deliberately not pooled. Different policy terms, occupations, covered expenses, claim thresholds, geographic mixes, and selection rules prevent a defensible combined average. Several sources also share CNA as an underwriting source, so they are not six independent samples of the entire market.

No national licensing-complaint incidence rate is estimated here. That would require a consistent count of unique matters and a matching exposure denominator, such as insured professional-years. Neither a collection of paid claims nor a count of disciplinary actions supplies that denominator. Disciplinary actions can also differ from complaints and from unique individuals.

The broad P&C data provide financial context. They are not BoardWise's addressable market. The reachable market for a service must instead be estimated from eligible matters, distribution access, adoption, service scope, and willingness to pay.

2 The financial setting

Broad property and casualty insurance

The following figures come from one NAIC annual series, preserving a consistent source rather than mixing preliminary estimates from different publishers. [1]

The NAIC also reports $86.003bn of loss expenses incurred and $252.164bn of underwriting expenses in 2025. These categories cover far more than administrative work suitable for AI. Their scale establishes context, not an automation opportunity of equivalent size. [1]

In simplified terms, a combined ratio below 100% indicates underwriting profitability before investment results; above 100% indicates an underwriting loss. Published ratios can use different premium denominators and include policyholder dividends. They should not be reconstructed by adding selected percentages without checking the source methodology.

BoardWise interpretation: a strong insurance balance sheet does not remove the incentive to improve service productivity. It may provide the funding to do so. Equally, a growing expense ratio is not proof of inefficient staff or failed technology. Product mix, commissions, growth investment, and cost inflation can all affect it.

Medical professional liability

An MPL Association article by Milliman actuaries analyzes 74 established specialty writers. For 2025 it reports more than $6.9bn in direct written premium, 1.7% premium growth, approximately $23.4bn in policyholders' surplus, and a combined ratio still above 100%. The composite excludes multiline carriers, smaller writers, and companies that exited or became insolvent. Its results therefore should not be presented as the full MPL market. [2]

This distinction matters commercially. An insurer may have substantial capital while still needing to improve the economics of individual coverage lines. A service provider should identify the relevant cost center and operational problem instead of relying on a general story about industry distress.

A further distinction is between loss adjustment expense, which includes handling and defending claims, and underwriting expense, which includes other costs of operating and acquiring insurance business. Savings from legal-file preparation will not automatically appear in the underwriting expense ratio. Classification depends on the actual service and accounting treatment.

Financial strength is not a measure of service quality

Neither a favorable combined ratio nor high capital demonstrates that policyholders receive timely, understandable help. Conversely, a difficult underwriting year does not establish poor claims conduct. A useful benchmark program measures financial performance and service performance separately, then studies their relationship.

For a professional facing an investigation, the relevant questions are immediate: What has arrived? What must be preserved? Which deadlines require confirmation? Who is qualified to advise? Who will pay? Those needs can exist before a carrier has completed a coverage review.

3 Licensing defense benchmarks across six professions

Published paid expense observations

These are selected public benchmarks, not quotations for representation and not actuarially comparable risk rankings. Each amount represents the expenses captured by its source, which may include items beyond attorney fees.

Changes within each published series

Calculation: percentage change = (later average ÷ earlier average − 1) × 100. These are changes between report datasets, not annual inflation rates. Comparability can still change within a publisher's series as insured populations and matter mix change.

What the benchmarks support

Paid regulatory defense is an identifiable expense category across multiple occupations. This makes it suitable for more detailed operational study: the initial intake, document collection, chronology preparation, professional review, communication, and continuing case management can be measured separately.

The opportunity is not established simply because an average bill is several thousand dollars. A substantial portion may reflect necessary legal judgment, hearings, experts, travel, or unusual facts. A service can only claim savings against work it actually changes. It must also account for the time professionals spend checking the service's output.

What the benchmarks do not support

The differences between professions do not show that one insurer is overcharging or one occupation is more likely to face discipline. They also do not establish that a fixed-price preparation service substitutes for a complete defense. Comparing a limited service fee with an entire multi-year representation expense would misstate the comparison.

The counselor data illustrate a denominator trap. Reported incidents and claims increased from 2,082 to 4,080 across five-year and six-year datasets. The publisher also identifies insured-population growth as a possible contributor. Those counts cannot establish a 96% increase in an individual counselor's risk. [5]

A second trap is treating a favorable outcome as a free process. The dental report places 80% of paid license-protection matters in its no-action category; average defense expense for that category was $3,882. [8] Necessary defense work can have value even when no sanction follows. No-action outcomes do not establish that every original complaint was frivolous or that defense could have been omitted.

4 Coverage and support are separate dimensions

Public product information shows that license protection already exists within parts of the professional-liability market. The service opportunity should therefore be evaluated against existing benefits, not against an assumption that professionals receive none.

These are public descriptions observed at the research cutoff, not a policy-form audit, a coverage opinion, or a ranking. Absence of a detail on a marketing page does not establish absence of coverage. Policy terms and endorsements control actual benefits.

A mature comparison should examine six dimensions: the event that triggers assistance; the point when an attorney becomes available; who advances payment; what expenses count toward the limit; who chooses counsel; and what help remains available when a matter is outside coverage.

The distinction between financing and service delivery creates several possible arrangements:

Covered-matter preparation. A service helps the professional organize information for appointed or selected counsel. The insurer continues to fund the covered defense. Counsel determines what material is useful and retains professional responsibility.

Support while coverage is being assessed. A carefully scoped service can help gather documents and identify questions for the carrier. It should not promise coverage, decide legal strategy, or encourage a professional to miss notification obligations while waiting for an automated answer.

Support for matters outside coverage. An insurer, association, or individual may purchase educational and organizational assistance. The economic rationale may be access or member satisfaction rather than a reduction in claims payments. An insurer cannot count money it was never obligated to spend as defense savings.

Association-funded access. A professional organization may subsidize support independently of insurance. This creates a distribution channel with different incentives and potentially different privacy expectations.

For each arrangement, the professional needs to understand who pays, who receives information, which services are included, and when an independent attorney should be engaged. A referral should not imply that a service is equivalent to legal representation.

5 The evidence on AI adoption and productivity

Adoption is substantial but definitions matter

The NAIC summarizes separate sector surveys as follows. These are historical survey observations, not synchronized October 2026 market shares. [12, 13]

The health survey was fielded November 2024 through January 2025 under 16 states' examination authority, using company-size or market-significance criteria. It is not a random sample of every U.S. insurer. Its current-use percentage cannot be compared directly with the other rows' broader use-or-exploration measures. [13]

Traditional machine learning used for fraud detection or pricing is also different from a generative system that drafts an explanation. Both differ from an AI-native service that coordinates an entire workflow. The useful purchasing question is not whether a vendor uses AI, but what work is performed, checked, delivered, and measured.

Direct evidence from an insurer

Lemonade's Q1 2026 investor presentation reports more than 50% claims automation and an approximately 6% loss-adjustment-expense ratio, defined there against gross earned premium. These are company disclosures, not an independently controlled comparison. Its product mix, claim complexity, reinsurance, and accounting differ from a licensing-defense operation. [22]

That example demonstrates that substantial automation is operationally possible in selected insurance settings. It does not establish that half of professional-regulatory matters can be automated, that the same expense ratio is achievable elsewhere, or that automation alone caused the company's results.

Experimental evidence has both upside and limits

The first two observations establish that task design and worker experience matter. The third measures expectations. Combining them into an assertion that all professional work will cost a fixed percentage less would erase the distinction between evidence and extrapolation.

For licensing support, a clean chronology with source references is a different task from deciding whether a consent agreement should be signed. A workflow can be effective at the former and inappropriate for autonomous handling of the latter. Evaluation must occur at the task level.

Purchasing rules are already changing

The Hartford's public panel-counsel guidelines require disclosure of generative-AI use, identification of the tool in corresponding billing entries, verification of outputs, and billing only for actual time spent. They also address confidentiality, records, and audits, and disallow passing through AI subscription or license fees under those guidelines. [23]

This is concrete evidence that at least one major purchaser has moved beyond experimentation to specifying how AI-assisted legal work should be delivered and billed. It is not evidence that every carrier has the same rules or that every matter is cheaper.

6 How AI native services could change insurance economics

Lowering the cost of useful preparation

A professional may arrive with notices, emails, credentials, employment records, and an incomplete account of events. Organizing that material can require repeated exchanges. A well-designed service could request missing items, preserve the difference between documentary evidence and recollection, and prepare a structured package for review.

The output must remain traceable. A date inferred by a model should not silently become an established fact. An allegation should not become an admission through an overconfident summary. A blank field should remain unresolved until the professional or an authorized reviewer supplies the information.

This creates a practical standard: measure the total time to produce an accepted, accurate work product, including corrections and professional review. Counting the seconds required to generate an initial draft would reward the wrong behavior.

Moving assistance earlier

Earlier help may prevent administrative mistakes, reduce repeated requests, and make an eventual attorney engagement more productive. These are plausible mechanisms, not established effects on disciplinary outcomes. Earlier access can also expose facts requiring more expensive intervention. A good escalation can raise short-term expenditure while protecting the professional and improving the integrity of the process.

Creating a new basis for competition

If policy limits and premiums are similar, practical support could become a point of differentiation. An association or insurer might compete on the time to a usable case summary, access to an appropriate attorney, clarity of communication, and continuity after a matter closes.

Whether users value those services enough to affect renewal must be tested. Satisfaction among people who used a benefit is not equivalent to improved retention across all policyholders. The evaluation needs an eligible comparison population and must account for pricing changes and other renewal factors.

Making customers more capable

The Hartford's 2026 benefits survey reports that 17% of 1,000 surveyed employed U.S. workers used AI during benefits enrollment. The survey was conducted in December 2025. This is adjacent evidence of AI-assisted consumer decision-making, not proof of adoption in professional-liability purchasing. [24]

Our forecast is that better assistance will make some professionals more willing to ask questions about exclusions, request written explanations, and compare available support. This may improve matching between needs and products. It may also increase service demand and disputes.

An insurer that defines success only as fewer payments could reject a service that helps people obtain benefits to which they are entitled. A more useful evaluation distinguishes administrative savings, valid benefit utilization, customer understanding, and actual errors in claims handling.

Changing the division of value

The party that performs a task faster does not automatically capture the savings. Under hourly arrangements, lower effort may reduce invoices. Under fixed fees, the provider may retain more of the benefit. In an internal operation, staff time may become additional capacity without a reduction in payroll. A customer may spend less time assembling records without any change to the insurer's paid expense.

Contracts should identify the intended benefit and how it will be measured. A fixed fee for a defined preparation package can be sensible when acceptance criteria are clear. Paying for a favorable board outcome would introduce different incentives and complications and is not assumed in this report.

7 An illustrative economic model

The following model is an original BoardWise scenario, not a forecast of BoardWise performance. Every input is an assumption. The $6,000 baseline is a round hypothetical cost for an eligible matter, not a pooled estimate from the profession-specific table. The service fee is not a BoardWise quote.

Per matter assumptions

Gross avoided expense = B × a × t × r.

Net benefit per handled matter = gross avoided expense − V − Q − O.

Existing professional review remains in the baseline unless genuinely reduced. Q represents additional review introduced by the new workflow. The model does not remove necessary counsel, hearings, or substantive legal work, and it assigns no reduction to indemnity payments or disciplinary sanctions.

The central case yields a 1.5% reduction in the assumed total expense. The upside case yields 8.7%; the conservative case raises expense by 3.9%. None of these is a claim about actual program performance.

At 75% cash realization, the break-even task improvement is 25%: $450 ÷ ($6,000 × 40% × 75%). At the central task and realization assumptions, the maximum break-even service fee is $390 after allowing $150 for additional review and variable administration. This is a procurement sensitivity, not a pricing recommendation.

Portfolio economics and adoption

Assume a program sees 1,000 potentially eligible matters annually, handles 60% through the service, and incurs $50,000 of annual fixed implementation, oversight, and integration costs. There are therefore 600 handled matters. At scale, the illustrated annual results become:

Under the central illustration, at least 556 handled matters are required to cover $50,000 of fixed cost, rounding up from 555.6. If only 30% of the 1,000 eligible matters use the service, annual net benefit becomes −$23,000. A useful tool can therefore fail commercially because too few people use it or because implementation is too expensive.

These figures exclude tax effects, financing, changes in claim frequency, retention value, and extraordinary losses caused by errors. Those effects require separate estimates. A serious error should not be hidden inside an average rework allowance.

Why unit savings may not reduce aggregate spending

Suppose, independently of the preceding model, unit cost falls 10% while utilization rises 15%. Total expenditure changes by 0.90 × 1.15 − 1 = +3.5%. More people receiving needed support can increase total spending even as each service becomes cheaper.

Likewise, a covered-matter program and an uncovered-matter benefit need separate business cases. The former may reduce existing defense expense. The latter purchases a new service and should be justified on access, member value, or another explicitly measured objective.

For underwriting impact, use the premium denominator appropriate to the expense classification. As a simplified scale illustration, $100,000 of net expense reduction against $100m of corresponding premium equals 0.1 percentage point. This calculation does not establish a change in an insurer's reported combined ratio without accounting reconciliation.

8 BoardWise forecasts through 2030

These are directional predictions. Confidence describes the strength of our judgment, not a statistical probability. Dates are evaluation horizons rather than promises. The forecasts are intentionally paired with observations that could weaken them.

1 Defined workflow services will gain ground in 2027 and 2028

Confidence: higher for preparation tasks; lower for autonomous consequential decisions. Buyers will increasingly ask for a completed, reviewable work product rather than access to an unrestricted chat interface. Document indexing, chronology preparation, missing-item checks, and status explanations have identifiable acceptance criteria.

The strongest evidence would be recurring paid contracts and renewals tied to these tasks. The prediction would weaken if review time consistently matches the time supposedly saved or if professionals prefer existing tools despite comparable access and training.

2 Legal service purchasing will become more explicit about AI

Horizon: 2027–2029. Confidence: higher. More purchasers will specify permitted tools, disclosure, verification, billing, and data handling. Some routine work will move toward fixed or capped prices, while complex work retains arrangements reflecting uncertainty and expertise.

Evidence would include revised panel guidelines, task-level budgets, and auditable invoices. The countercase is that rate increases, staffing changes, and matter complexity absorb productivity gains, leaving total paid costs unchanged. The Hartford's existing guidelines are an early example, not proof of market-wide convergence. [23]

3 Service quality will become a more visible insurance differentiator

Horizon: 2027–2030. Confidence: moderate. Selected professional-liability programs and associations will add structured licensing support or make existing support easier to access. Buyers will compare the practical experience of obtaining help alongside premium and limits.

The test is whether benefits reach users and improve measured service or renewal outcomes. A branded portal with little usage would not confirm the prediction. Neither would bundling a feature that replaces a more valuable existing benefit.

4 Easier access will expand demand as well as reduce effort

Horizon: 2027–2029. Confidence: moderate. Some professionals who previously deferred assistance will seek it earlier. Associations may receive more requests after publicizing support. Increased utilization could offset savings and expose previously unmet need.

Evidence would include earlier reporting, higher take-up among eligible professionals, and a broader range of case complexity. The prediction would weaken if usage remains concentrated among people already receiving comparable assistance.

5 Independent assistance will make coverage gaps more visible

Horizon: 2027–2030. Confidence: moderate. Tools that help professionals organize correspondence and formulate questions may expose mismatches between expected protection and actual policy terms. Insurers may respond with clearer communication, revised benefits, or new service partnerships.

This forecast does not assume wrongful denial. A legitimate exclusion can still create an unmet support need. Relevant measures include user understanding, documented coverage inquiries, corrected errors, and the proportion of referrals successfully connected to suitable help.

6 Distribution partners may capture substantial value

Horizon: 2028–2030. Confidence: moderate. Associations, brokers, program administrators, and insurers already connected to professionals may become effective distributors of AI-native support. Embedded access could be easier to find at the moment of need than a standalone consumer product.

The countervailing possibility is direct AI-assisted discovery: professionals may ask their preferred assistant for help and bypass established channels. Evidence should track acquisition cost, referral activation, conversion, and retention by channel rather than website traffic alone.

7 Governance and integration will constrain adoption more than drafting speed

Horizon: 2027–2030. Confidence: higher. In consequential workflows, buyers will demand reliable escalation, source traceability, data controls, and accountability. A fast demo will be insufficient when the operation must manage real records, exceptions, and disputes.

The test is whether procurement and renewals depend on these capabilities. The forecast would weaken if buyers sustainably accept low-control systems without increased errors, remediation costs, or loss of trust. This is a prediction about purchasing behavior, not a claim that any one compliance checklist guarantees safety.

8 Better service evidence will emerge before credible loss reduction evidence

Horizon: 2027–2030. Confidence: higher. Organizations should be able to measure preparation time, completeness, corrections, and human effort before they can establish causal effects on indemnity, sanctions, or total loss ratios.

Claims of reduced losses will require adequate follow-up and comparable populations. A pilot can demonstrate useful operational results without having enough cases or observation time to support a disciplinary-outcome claim. The prediction would be challenged by large, well-designed evaluations producing credible outcome evidence sooner.

9 Where the thesis could fail

Human review may consume the gain. A reviewer who must reconstruct the entire matter to detect subtle errors may save little time. Evaluate the complete workflow, including user corrections, not just model output.

The hardest cases may dominate economics. A minority of complicated matters can account for a large share of expense. A service that improves routine tasks may have little effect on those cases. Report the mean, median, and upper-tail costs separately.

Generic tools may narrow differentiation. If document summaries become widely available at low cost, a specialized provider must demonstrate better workflow fit, reliable sources, appropriate escalation, and distribution. Access to a capable foundation model alone is not a lasting advantage.

Professional rules may limit service scope. Educational support, clerical organization, individualized legal advice, claims adjusting, and insurance distribution can involve different obligations. A business should obtain jurisdiction-specific review of its actual activities, contracts, and representations. A disclaimer does not determine the legal character of the service.

Privacy commitments may conflict with operational expectations. A low-retention service may appropriately avoid storing case content, while a buyer expects extensive audit records. The parties must design compatible evidence and retention arrangements before contracting. Identifiable case content should not be retained merely to create a marketing dataset.

Trust can be damaged by divided loyalties. A professional may reasonably worry that candid disclosures could reach an employer or insurer. Agreements should explain information access, permitted uses, and any limits to confidentiality. Sponsorship must not be mistaken for professional independence or attorney-client privilege.

AI may create correlated errors. Many cases can share a faulty template, an outdated rule, or a model behavior. Oversight must detect repeated patterns as well as isolated incidents. An average quality score can conceal a serious failure concentrated in one state or profession.

A service can improve access without creating insurer savings. This is not automatically failure, but the buyer must choose the correct objective. Changing the evaluation after a pilot to whichever metric improved would undermine credibility.

10 Governance requirements for credible deployment

The NAIC's model AI bulletin sets expectations around insurer governance, documentation, and consumer-impacting AI use, including third-party systems. Its adoption by the NAIC is not itself a uniform law in every state; applicable state requirements must be checked. The NAIC maintains a state implementation map, including an October 8, 2026 version reviewed for this report. [17, 18]

NIST's generative-AI profile provides a voluntary risk-management reference addressing risks such as unreliable outputs, privacy, and information integrity. It is not a certification of a vendor. [19] ABA Formal Opinion 512 discusses lawyers' obligations involving competence, confidentiality, communication, and fees when using generative AI; applicable professional rules remain jurisdiction-specific. [20]

For this service category, our proposed minimum evaluation covers:

These are proposed buying criteria, not representations that BoardWise or any cited vendor has independently passed them. Verification should be proportionate to the service's actual authority. A tool that organizes records creates different risks from one authorized to transmit a response, make a representation, or execute a transaction.

11 A practical pilot and benchmarking standard

Define the population before launch

A pilot should specify professions, states, notice types, matter stages, language needs, coverage status, and permitted service tasks. Cases needing urgent legal intervention should follow an established escalation path; excluding them from routine automation must not mean withholding help.

Track the entire funnel: referred, eligible, offered, started, completed, escalated, and followed up. Report why people were excluded or did not complete the workflow. A completion-only result can conceal access barriers and selection bias.

Separate operational and professional outcomes

Use a comparison that can answer the question

Random assignment may be appropriate for eligible, nonurgent preparation tasks when everyone retains ordinary support and necessary counsel. If randomization is impractical, use a contemporaneous comparison matched on profession, state, allegations, stage, and complexity. Disclose residual differences rather than implying that matching eliminates all bias.

Measure outcomes for everyone assigned or offered the service, not only enthusiastic adopters. Analyze completion and usage separately. Independently review a defined sample for quality, including cases with poor results. Use the same cost definitions in both groups.

A 90-day period may be useful for testing usability and preparation workflows. It is not automatically enough to assess a board matter's outcome. Set follow-up intervals according to the actual process, retain open cases in reporting, and avoid treating unresolved matters as successes.

Sample size should follow the selected outcome, baseline variability, desired precision, and acceptable safety threshold. A convenient round number of cases is not a statistical justification. Report uncertainty intervals where appropriate and disclose small subgroups rather than making precise claims from them.

Predefine the decision to expand

Before launch, agree on minimum quality, acceptable error severity, escalation performance, cost measurement, and the conditions for pausing the program. A favorable average cost result must not override a serious deterioration in accuracy or access to counsel.

For renewal, require evidence that users actually receive the service, professional reviewers accept its output, and the complete cost is consistent with the stated objective. For expansion into another profession or state, reassess the workflow instead of assuming that success transfers unchanged.

12 Implications for insurers and the professional support market

For insurers, the immediate opportunity is to identify bounded service tasks and contract for measurable outcomes. A broad transformation initiative can begin with a small operational problem, but its business case must survive the full cost of oversight and integration.

For brokers and program administrators, the opportunity is clearer benefit communication and easier access to relevant help. Product comparisons should explain what happens after a notice arrives, not simply list a defense limit.

For professional associations, the opportunity is access. A member benefit may be worthwhile even where an insurer cannot demonstrate direct savings. Evaluation should reflect the association's purpose and preserve members' confidence about who sees their information.

For attorneys, the opportunity is receiving better-organized material and spending a larger share of effort on judgment, advice, and advocacy. The risk is that low-quality preparation creates more review work or encourages inappropriate self-representation. Attorney involvement in workflow design is therefore commercially useful as well as professionally important.

For AI-native service providers, the challenge is to demonstrate a dependable service rather than a persuasive demonstration. Useful evidence will include accepted work products, transparent boundaries, real adoption, accountable escalation, and reproducible cost measurement.

The role BoardWise is seeking to build

BoardWise's public insurance-partnership offering focuses on regulatory education, licensing-matter preparation, resource navigation, and exploration of scoped pilots. Its stated role is educational and organizational assistance, with independent attorney engagement available separately. [21]

Our strategic thesis is that professionals should be able to obtain structured support before they are overwhelmed by a process they do not understand. Insurers and associations can help distribute that support, while independent professionals retain control over their decisions and access to qualified advice.

That thesis must be evaluated through evidence. This report does not claim that BoardWise has reduced an insurer's combined ratio, replaced a complete legal defense, or caused better disciplinary outcomes. The next stage is to measure what changes in actual workflows and publish results with the same care used to describe the public benchmarks.

Organizations interested in evaluating a defined professional-support pilot can visit BoardWise insurance partnerships. Initial inquiries should describe the proposed use case without including sensitive case information.

References and source notes

All sources accessed October 26, 2026. Numbered references identify original publications or the original organization's public description. Numerical entries retain their source populations; no unpublished insurer data are implied.

1. NAIC. U.S. Property and Casualty and Title Insurance Industries — 2025 Full Year Results. P&C summary, PDF page 1. Basis for the annual financial table. Source
2. Eric J. Wunder and Zachary A. Fischer, MPL Association. Industry Update: 2025 Financial Results — Capitalization Levels Soar Despite Underwriting Losses. 2026 article; 74 established specialty-writer composite. Source
3. CNA and NSO. Nurse Professional Liability Claim Report, 5th Edition. 2025; PDF pages 20–21, methodology and Figure 18. Source
4. CNA and NSO. Nurse Practitioner Spotlight: Defending Your License. Companion to the 2022 fifth-edition report; PDF page 1, paid-matter comparison. Source
5. CNA and HPSO. Counselor Professional Liability Exposure Claim Report, 3rd Edition. 2024; PDF pages 5–6 and 24–25, methodology and Figure 21. Reported-incident counts differ from the paid-matter denominator. Source
6. CNA and HPSO. Physical Therapy Professional Liability Claim Report, 5th Edition. 2026; PDF pages 5 and 20–21, methodology and Figure 21. Source
7. HPSO and CNA. Pharmacist Professional Liability Exposure Claim Report, 3rd Edition. Publisher's key-findings page, including comparison with the 2018 dataset. Source
8. CNA and Dentist's Advantage. Dental Professional Liability Claim Report, 3rd Edition. 2025; PDF pages 4 and 10, methodology, defense expense and board outcomes. Source
9. NSO. Protect Your Professional License, Protect Your Livelihood. Public coverage explanation; not an issued policy. Source
10. HPSO. Everything You Need to Know About Your Coverage. Public coverage explanation; profession-specific terms require verification. Source
11. CM&F. Registered Nurse Malpractice Insurance. Public product description listing license-defense coverage. Source
12. NAIC. Insurance Topics: Artificial Intelligence. Summary of separately conducted sector surveys. Used for auto, home and life use-or-exploration figures. Source
13. NAIC. NAIC Survey Reveals Majority of Health Insurers Embrace AI. May 20, 2025; methodology and current-use result. Source
14. Erik Brynjolfsson, Danielle Li and Lindsey Raymond. Generative AI at Work. Revised abstract reports 5,172 agents and a 15% average productivity gain; earlier working-paper versions reported different rounded results. Source
15. Fabrizio Dell'Acqua and colleagues. Navigating the Jagged Technological Frontier. Organization Science, published online March 17, 2026; experiment using GPT-4 in 2023. Abstract and results discussion, including PDF page 12 for the correctness difference. Source
16. Thomson Reuters. The AI Adoption Reality Check. June 26, 2025; Future of Professionals survey release. The time-saving figure is a respondent expectation. Source
17. NAIC. Model Bulletin: Use of Artificial Intelligence Systems by Insurers. Adopted December 2023. Model expectations; state implementation must be checked separately. Source
18. NAIC. Implementation of NAIC Model Bulletin: Use of Artificial Intelligence Systems by Insurers. Map marked October 8, 2026. Linked file may be updated after publication. Source
19. NIST. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1, July 2024. Voluntary risk-management reference. Source
20. American Bar Association. ABA issues first ethics guidance on a lawyer's use of AI tools. July 29, 2024; summary of Formal Opinion 512. Source
21. BoardWise. Professional Liability Insurance Partnerships. Publisher's own service positioning; not independent validation. Source
22. Lemonade. Investor Presentation, updated March 31, 2026, PDF page 10. Self-reported automation and LAE measures; company-specific definitions. Source
23. The Hartford. Claims Panel Counsel: Use of Artificial Intelligence. Public operational and billing guidelines, reviewed at the research cutoff. Source
24. The Hartford. Amid Rising Cost of Living, AI Emerging as Benefits Decision-Making Tool for Gen Z. May 7, 2026; survey methodology and 17% adoption observation. Source