Why the 'future of underwriting' question is being asked differently in 2026
The question 'where is underwriting heading' has two very different answers depending on who is asking. Vendors selling workbenches promise replacement of underwriters in 36 months. The CUOs and Boards I work with at $500M-$5B GWP P&C carriers are planning very differently: they are encoding the institutional knowledge of senior underwriters before retirement, then layering AI on top of working rules. That sequence, codify first and augment second, is the actual direction of the next three years.
The evolution from manual to digital to AI-augmented underwriting
U.S. P&C underwriting has moved through three eras in the last 30 years. Manual underwriting (pre-2000) ran on paper applications, mainframe rating systems, and senior underwriter judgment with little system support. Digital underwriting (2000-2020) added rating engines bolted to PAS, electronic submissions, basic rules engines, and document management. AI-augmented underwriting (2020-onward) layers machine learning scoring, generative AI document extraction, and explainability on top of the digital foundation.
What gets lost in the marketing is that the AI-augmented era still requires the digital foundation. Carriers that tried to skip directly from manual workflows to ML-driven underwriting consistently produced bad models on bad data, then ran into NAIC scrutiny. The carriers I have advised that succeeded with AI augmentation built strong rules engines first, then added ML scoring inside the rule boundaries. The sequence matters.
Five shifts shaping underwriting workbenches through 2028
Generative AI moving from extraction to drafting
In 2023-2024, generative AI in underwriting was used primarily for extracting structured data from messy broker submissions: ACORD forms, loss runs, schedules of values. By 2026, generative AI is also drafting decline letters, surcharge explanations, and regulator-facing audit narratives based on the underwriter's decision. By 2028, expect GenAI to draft full coverage analyses and policy comparisons that the underwriter then edits and approves. The underwriter's role becomes editor and authority, not author.
Explainability as a hard regulatory requirement
The NAIC AI Model Bulletin moved from guidance to enforced policy in 24 jurisdictions between late 2024 and August 2025. The NAIC AI Systems Evaluation Tool pilot running in 12 states through September 2026 is examining carriers' AI governance programs. By 2028, every state DOI will examine AI model documentation as part of standard market conduct examinations. Carriers that adopted explainability layers (SHAP, counterfactuals) voluntarily in 2024-2025 are in a strong position. Carriers that ran black-box scoring are facing retrofit costs and, in some cases, injunction risk.
No-code rules authoring becoming standard
The CUO-authored rule, deployed in 24 hours instead of 4 months through IT, is no longer a competitive advantage; it is becoming table stakes. By 2028, expect every credible underwriting workbench vendor to support CUO-controlled rules authoring with simulation, version control, and governance workflow. The differentiator will shift to integration depth, AI quality, and audit-trail completeness.
Real-time portfolio exposure replacing monthly snapshots
Most mid-tier P&C carriers still see exposure on a monthly cadence: month-end reports summarized by line of business and geography. By 2028, real-time portfolio dashboards showing accumulation by geography, catastrophe zone, and reinsurance treaty will be a baseline expectation for CUO oversight. This shift is being driven by the post-2023 wildfire and hurricane seasons exposing accumulation that monthly reporting missed.
Specialty and MGA lines closing the technology gap
Specialty lines (cyber, ocean marine, aviation, equine, agriculture) and MGA-written books historically lagged personal and small commercial in underwriting technology, because vendor workbenches were configured for high-volume standardized risks. Modern workbenches with configurable data models are closing that gap. By 2028, specialty carriers and MGAs that have invested in workbench technology will have decision-time and audit-quality advantages over peers still running Excel-and-email workflows.
AI and predictive analytics in underwriting: what is real in 2026
Cutting through the AI marketing, four use cases are now production-ready at mid-tier P&C carrier scale, based on deployments I have advised on or reviewed.
Document extraction is the most mature. Generative AI plus OCR now handles ACORD forms, loss runs in arbitrary carrier formats, schedules of values in inconsistent Excel layouts, and unstructured email submissions at production accuracy. The technology is boring now, which is the right state for it.
Risk scoring is production-ready for high-volume standardized lines (personal auto, homeowners, small commercial property). For specialty and complex commercial, scoring models exist but typically run as decision support: they recommend, and the underwriter decides. Carriers attempting fully automated scoring on complex risks consistently report higher loss ratios than augmented workflows.
Predictive analytics for portfolio management (catastrophe accumulation, adverse selection patterns, broker performance) is also production-ready. The bottleneck is usually data quality and integration, not the analytics. The carriers I work with that saw the most value here invested 6-9 months in data remediation before any analytics deployment.
Fraud and adverse selection detection at quote (pre-bind) is still maturing. Models exist; production accuracy on specialty lines remains inconsistent. Expect significant progress through 2028 as more carriers share anonymized model training data through industry consortia.
Regulatory direction: NAIC AI Bulletin and what comes next
NAIC AI Model Bulletin, adopted in 24 jurisdictions as of August 2025, set the current regulatory baseline. It requires insurers to maintain a written AI Systems Program covering governance, risk management, internal controls, model documentation, third-party model oversight, and prevention of unfairly discriminatory outcomes.
Three regulatory directions are visible through 2028. First, the NAIC AI Systems Evaluation Tool pilot (currently running in 12 states through September 2026) will likely become a standard market conduct examination component nationwide. Second, state-specific AI rules will continue layering on top (Colorado, New York, Connecticut have moved further than NAIC baseline). Third, federal-level AI oversight of insurance remains uncertain but probable in some form by 2028.
The operational implication: carriers planning workbench investments in 2026-2027 should build for the regulatory environment of 2028, not 2024. That means explainability layers configured at deployment, per-decision audit trails as a default, and model governance documented in a form examiners actually use. The detail on what this looks like at the carrier level is in our regulatory compliance article.
What this means for mid-tier P&C carriers planning 2026-2028 roadmaps
From the Board sessions I sit in on, three planning principles consistently separate carriers building durable workbench positions from carriers cycling through vendors.
Codify before you automate. Encoding the institutional knowledge of senior underwriters into rules and decision frameworks should come before ML scoring layers. The carriers that did this in 2023-2024 had a working foundation by the time NAIC enforcement tightened in 2025.
Build for the 2028 regulator, not the 2024 examination. Explainability, audit trail, and model governance are cheaper to design in than to retrofit. Retrofits I have seen ran 18-36 months and 7-figures.
Pick a workbench architecture that does not lock you into one vendor's data model. The carriers that bought monolithic vendor platforms in 2010-2015 are facing 24-36 month exit projects. Modern PAS-agnostic workbenches (workbench layered on Guidewire, Duck Creek, or Majesco) preserve optionality.
For the foundational view of what an underwriting workbench actually is, see our underwriting workbench pillar guide. For the question on AI and senior underwriters specifically, see will AI replace underwriters. And for the operational diagnostic on readiness, see 5 signs your carrier needs a workbench.
FAQ
What is the future of underwriting in 2026-2028 for U.S. P&C carriers?
The direction for U.S. P&C carriers through 2028 is AI-augmented underwriting, not AI-replaced underwriting. Workbenches will increasingly handle document extraction, rule application, and risk scoring recommendations, while underwriters retain authority over consequential bind, decline, and pricing decisions on specialty and complex risks. Five specific shifts will define the period: generative AI moving from extraction to drafting, explainability becoming a hard regulatory requirement, no-code rules authoring becoming table stakes, real-time portfolio exposure replacing monthly snapshots, and specialty/MGA lines closing the technology gap with personal lines.
How will AI change underwriting jobs by 2028?
AI will continue to compress the time underwriters spend on document processing (currently 60-70% of a typical day in legacy workflows) and shift senior underwriter capacity toward portfolio strategy, broker relationships, and complex risk evaluation. The carriers I have advised that framed AI deployments as augmentation kept their senior talent and saw 2-3x productivity gains in policies per FTE. The carriers that framed deployments as replacement consistently lost senior underwriters and inherited the regulatory and operational consequences. Through 2028, expect headcount in routine underwriting to decline modestly while specialty and complex underwriting headcount stays flat or grows.
What is AI-driven underwriting and how does it differ from traditional underwriting?
AI-driven underwriting layers machine learning scoring, generative AI document extraction, and explainability tooling on top of a rules-engine foundation. Traditional underwriting relies on rating engines, guideline documents, and senior underwriter judgment without ML support. The difference is not the elimination of underwriter judgment; it is augmentation. AI handles routine extraction and scoring; the underwriter handles consequential decisions, broker relationships, and complex risks. Production-ready use cases in 2026 include document extraction, risk scoring for high-volume standardized lines, and portfolio analytics.
How does the NAIC AI Model Bulletin affect underwriting workbench plans for 2026-2028?
NAIC AI Model Bulletin, adopted in 24 jurisdictions as of August 2025, requires a written AI Systems Program covering governance, model documentation, bias testing, and third-party model oversight. The NAIC AI Systems Evaluation Tool pilot in 12 states through September 2026 will likely become a standard examination component nationwide. Carriers planning workbench investments should design for the 2028 regulatory environment from day one: explainability layers, per-decision audit trails, and model governance configured at deployment. Retrofitting these elements after the fact has cost carriers I have advised 18-36 months and seven-figure budgets.
Will predictive analytics replace traditional underwriting models?
Predictive analytics will supplement, not replace, traditional rating and underwriting models in the 2026-2028 horizon. Rules engines continue to define hard eligibility and pricing guardrails because regulators require explicit, documented decision logic for premium impacts. Predictive models score risks within those boundaries, surfacing patterns rules cannot capture. The combined output (rule path plus model features plus underwriter judgment) is what state DOIs increasingly examine. Pure predictive-only scoring is rarely deployable in regulated lines, which is why the carriers I work with consistently keep rules as the foundation.
What should mid-tier P&C carriers prioritize in their 2026-2028 underwriting roadmap?
Three priorities consistently show up in the Board sessions I sit in on. First, codify the institutional knowledge of senior underwriters into rules and decision frameworks before adding ML scoring. Second, design for the 2028 regulatory environment from day one with explainability, audit trail, and model governance built in. Third, pick a workbench architecture that does not lock you into one vendor's data model: PAS-agnostic workbenches layered on Guidewire, Duck Creek, or Majesco preserve optionality. Carriers that get all three right typically see 3-5 points of loss ratio improvement within 12-18 months of deployment.
Talk to Decerto about Higson
If you are a Chief Underwriting Officer or VP Underwriting at a U.S. P&C carrier in the $500M-$5B GWP range, and your current underwriting workflow runs on a mix of PAS screens, Excel rating sheets, and email-based broker submissions, that is the exact pattern Higson was built to address. Higson layers on top of Guidewire PolicyCenter, Duck Creek Policy, or Majesco without forcing a PAS replacement, and ships a CUO-controlled rules engine that cuts rule deployment from a 4-month IT backlog to 24 hours.
My take, after 20 years on the carrier side: Higson is not the right fit for $5B+ enterprise carriers running multi-region multi-currency books across 30+ jurisdictions. Guidewire is. Higson is built for U.S. P&C carriers in the $500M-$5B GWP range that need rules-engine control and audit-grade documentation without a 24-month PAS replacement.
Sources
- National Association of Insurance Commissioners (NAIC). (2023, December 4). Model Bulletin on the Use of Artificial Intelligence Systems by Insurers
- National Association of Insurance Commissioners (NAIC). Implementation of NAIC Model Bulletin: Use of Artificial Intelligence Systems by Insurers (state adoption tracker).
- National Association of Insurance Commissioners (NAIC). AI Systems Evaluation Tool Pilot: Pilot Project Background.
- New York State Department of Financial Services. (2024, July 11). Insurance Circular Letter No. 7 (2024): Use of Artificial Intelligence Systems and External Consumer Data and Information Sources in Insurance Underwriting and Pricing.
- Colorado Division of Insurance. (2024, October 18). Notice of Adoption: New Bulletin B-10.004 Concerning the Quantitative Testing Reporting Requirement for Life Insurers that Use External Consumer Data and Information Sources.
- Baker Tilly. (2026). The Regulatory Implications of AI and ML for the Insurance Industry.
Subscribe to receive the latest blog posts to your inbox every week.
