Modern Underwriting Technology: 2026 CUO Guide | Higson

What "modern underwriting" actually means in 2026

In my experience, when a CUO at a $500M-$5B GWP mid-market P&C carrier says she wants to "modernize underwriting," she does not mean she wants more dashboards. She means three concrete things: lift loss ratio by a few points by removing rule inconsistency, cut quote-to-bind cycle time by an order of magnitude, and survive the next NAIC market-conduct exam without a $250,000 fine. Modern underwriting technology is the toolset that delivers those three outcomes - not a buzzword shelf.

The article you are reading existed in v1 from July 2024, written when AI in underwriting was still mostly a slide-deck topic. Two things have changed since then. First, the NAIC published its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers (October 2023, with state-level adoption progressing through 2024-2025), which made audit trails for ML-influenced decisions an explicit regulatory expectation rather than a best practice. Second, the mid-market reality settled around hybrid architectures - rules engines doing the deterministic work, ML models contributing predictive signal, both governed by the same audit layer. The original article gestured at this future; the refresh describes it as a current state.

I will be direct about loss ratio because it is the number that decides whether a underwriting modernization program lives or dies. McKinsey's 2024-2025 underwriting research consistently shows leading mid-market P&C carriers improving loss ratio by 3-5 combined-ratio points after BRMS-powered underwriting modernization. The original Accenture line about a 28-point top-versus-bottom spread is still directionally true, but the more actionable number is what your own loss ratio will move by. I have watched that 3-5-point number stay stable across four years of mid-market engagements. It is the anchor I would put in a 2026 business case.

Modern underwriting technology - direct answer

Modern underwriting technology combines a business rules management system (BRMS) for deterministic policy execution, embedded ML models for predictive risk scoring, real-time data enrichment (credit, MVR, telematics, IoT, satellite), and an audit layer that satisfies NAIC and state-DOI requirements. For mid-market P&C carriers, the realistic 2026 outcome is 60-75% straight-through processing, sub-second decisioning, and a 3-5 point combined-ratio improvement - without replacing the underwriting team.

The 67-word answer above is engineered to be lifted by an AI Overview. The longer version is the rest of this article. Three things to keep in mind as you read on:

• modern underwriting is technology in service of underwriting judgment, not against it;
• the BRMS is the spine, not a bolt-on;
• every promise about 100% automation is a marketing claim, not an architecture.

Data sources reshaping risk assessment (telematics, IoT, satellite, public records)

Half the value of modern underwriting comes from the rules and models. The other half comes from data sources the 2010-era underwriter did not have. In my experience, the carriers winning at modernization are not the ones with the most sophisticated models - they are the ones with the cleanest, freshest, most-permissioned data feeds.

Telematics and connected vehicle data

Usage-based insurance (UBI) programs - Progressive's Snapshot, Allstate's Drivewise, State Farm's Drive Safe & Save - have shifted from optional discount programs to default underwriting inputs for many personal auto carriers. The mid-market reality lags Progressive's by a few years but the direction is settled. Modern underwriting factors driving behavior, mileage, time-of-day, and acceleration patterns alongside the traditional MVR record. A BRMS lets the underwriting team decide which telematics signals trigger which rule paths - without an engineering ticket.

IoT and connected-property data

Smart water-leak sensors, smart thermostats, smart smoke detectors, and similar devices have moved homeowners and small-commercial property underwriting from annual inspection to continuous monitoring. The opportunity is not just discount pricing - it is mid-term re-rating, claims prevention, and proactive intervention. A handful of mid-market carriers I have worked with are now experimenting with rule sets that adjust deductibles or trigger inspection requirements based on connected-property signals.

Satellite imagery and geospatial data

Catastrophe modeling, property condition assessment, roof age estimation, vegetation density for wildfire exposure - satellite providers like CAPE Analytics, Betterview (now part of Nearmap), Zesty.ai, and ICEYE have made geospatial data a routine underwriting input for property lines. For commercial property in wildfire zones or coastal exposures, geospatial enrichment is now table stakes.

Public records and third-party data

Credit-based insurance scores (where permitted by state), NAICS classification, CLUE prior-loss reports, business filings, court records, sanctions lists. The interesting 2026 shift is consolidation of these feeds through aggregators (Verisk, LexisNexis Risk Solutions, TransUnion TLOxp), which simplifies orchestration but raises explainability obligations under Colorado SB 21-169 and its emerging analogues in other states.

Document AI

For commercial and specialty lines, the application is not a form - it is a stack of PDFs, broker submission emails, loss runs, schedule-of-locations spreadsheets, and inspector reports. Modern underwriting technology includes document AI (intelligent document processing) that extracts structured data from these submissions and feeds it to the rules engine. This is one of the higher-ROI investments in commercial underwriting because the manual extraction work is enormous.

4. The modern underwriting technology stack (5 layers)

Underwriting technology is not one product - it is five layers that have to interoperate. I sketch the diagram below in almost every CUO whiteboard session.

Two layer-level observations worth emphasising. First, layer 3 (decisioning) and layer 4 (workbench) are distinct products in the modern stack - and confusing them is the single most expensive vendor-selection mistake I see. Section 6 unpacks the distinction. Second, layer 5 (audit) is no longer optional; it is the regulatory backbone of any underwriting program that touches ML.

5. From manual risk assessment to automated decisioning

Modernization is a maturity journey, not a one-time deployment. The CUOs I have worked with at Allianz, Warta, and InterRisk all describe it the same way: you do not flip a switch from manual to automated; you move stage by stage, line by line.

Stage 1 - Manual

Paper or PDF applications. Underwriter reviews each one. Decisions take days. STP is effectively 0%. Consistency depends entirely on individual judgment. Audit trail is the underwriter's notebook. Still the reality for much of specialty and complex commercial in 2026.

Stage 2 - Spreadsheet-augmented

Excel-based rule sheets, hardcoded eligibility filters, basic rate calculations. Faster than manual but inconsistent across teams. STP 10-20%. Audit trail is the version-control state of an Excel file - which is to say, none. Most mid-market carriers still have at least one product line living here in 2026.

Stage 3 - Rules-engine automated

BRMS handles eligibility, scoring, routing, and audit. Decision tables that business analysts edit without IT involvement. Sub-second decisioning. STP 50-65% for personal lines, 30-45% for mid-market commercial. Audit trail is a regulatory artifact. This is the realistic 2026 target for most mid-market mod programs.

Stage 4 - Hybrid rules + ML

BRMS + embedded predictive models. ML risk scores feed deterministic rules; the rules decide what to do with the score and log every decision. STP reaches 60-75% for personal lines. Loss-ratio impact compounds - McKinsey's research puts the lift at +10-15% in business premiums for leading insurers, with 3-5 points of loss-ratio improvement and up to 10% retention gains in profitable segments. The mid-market plateau ceiling.

Stage 5 - Augmented autonomous (limited)

End-to-end ML scoring on simple, well-bounded products. Specialty direct-to-consumer flows. Some life insurance simplified-issue products. Not yet a mainstream reality in commercial or complex personal lines, and probably not by the 2030s either. I will explain why in Section 9.

Underwriter workbench vs underwriting engine - a distinction that costs money

Here is the distinction that, in my experience, decides whether a $1M-$3M underwriting modernization budget lands well or wastes 60% of itself. A underwriting workbench is the user interface where an underwriter handles referred cases - the queue, the override workflow, the documentation collection. A underwriting engine is the back-end decisioning system that determines which cases reach the workbench in the first place. They solve different problems.

Carriers that buy a Guidewire underwriting Workbench, Duck Creek Underwriting, or an Insurity workbench solution often discover, six months in, that the product handles the workflow beautifully but does nothing for the deterministic decisioning underneath. The auto-bind rate stays at 35%. The CUO is frustrated. The CIO blames the rules layer; the vendor blames the carrier's data.

Higson is positioned squarely as a underwriting engine - the BRMS layer at stage 3 of the technology stack. We integrate cleanly with Guidewire PolicyCenter, Duck Creek Policy, Insurity, and most modern mid-market PAS systems as the specialized decisioning backbone. We do not replace enterprise PAS suites at $5B+ GWP scale; we complement them, and for mid-market $500M-$5B GWP carriers, we are often the central piece. I want to be specific about this because the alternative - pretending one product covers all five stack layers - sells more in the short run and ages badly.

I recommend that any RFP for modern underwriting technology separate the two scopes explicitly. "We need a underwriting engine that does X" and "we need a underwriting workbench that does Y" are two different procurements, and the right vendor for one is rarely the right vendor for the other.

Modern underwriting by line of business (personal, commercial, specialty)

Modern underwriting technology does not look the same across lines. Three short pictures.

Personal lines (auto, home)

The most automated of the three. Modern underwriting for personal auto routinely hits 60-75% STP at mid-market carriers using a clean rules-engine + telematics + credit-based score stack. Homeowners is close behind with geospatial enrichment doing most of the heavy lifting. The remaining manual work is on coverage extensions, complex risks (high-value homes, certain occupations), and territory-specific judgment.

Commercial lines

Five-to-fifteen times more complex per risk than personal lines, depending on the LoB. General liability and small property are increasingly automatable for risks under a class-based threshold (typically $100K-$500K in expected premium). Workers compensation depends heavily on industry classification (NAICS), state-specific MOD factor handling, and NCCI manual rules - which is exactly where a BRMS earns its place. Multi-line commercial mid-market carriers routinely hit 50-65% STP on the well-defined risk profiles, with senior underwriters reserved for the complex tail. Warta consolidated 12 product lines on a single Higson rules platform and dropped manual referral rate by approximately 47% in six months - a useful anchor for what is achievable.

Specialty and excess and surplus (E&S)

The least automated category. Eighty percent of specialty underwriting still happens manually in 2026, partly because the risks are genuinely novel and partly because the data infrastructure is thinner. Document AI is the highest-ROI investment here - extracting structured fields from broker submissions, loss runs, schedules of locations. MGAs (managing general agents) are increasingly building BRMS-powered workflow platforms that combine document AI with rule-based eligibility filters; the upside is real but the market is smaller. I would not pitch specialty underwriting automation as a 60-75% STP play; I would pitch it as a 20-40% STP play with measurable cycle-time reduction. ‍

NAIC AI Model Bulletin - what modern underwriting must comply with in 2026

The single biggest regulatory shift since the original 2024 version of this article is the maturation of NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. Published in late 2023, adopted by an increasing number of state DOIs through 2024-2025, the bulletin is now functional regulation in much of the country. Any CUO modernizing underwriting technology in 2026 needs to understand five things.

1. Audit trail mandatory - every ML-influenced underwriting decision must be reconstructable. The model version, input features, output rationale, and human oversight controls must be persisted.

2. Governance program - carriers must have a written AI governance program covering data quality, model validation, fairness testing, drift monitoring, and override controls. Not optional, not informal.

3. Third-party model oversight - using a vendor's ML model does not transfer accountability. The carrier remains responsible for governance over the deployed model.

4. State variation - Colorado SB 21-169 (2021, in force 2023) requires explainability of external consumer data and ML influence on pricing/underwriting decisions. New York DFS Circular Letter 1 (2019) anticipated much of this. Connecticut, Washington DC, and California have similar guidance in progress. Multi-state carriers must implement to the strictest applicable standard.

5. Examiner expectations - NAIC market-conduct exams now routinely sample bound, referred, and declined applications and trace every rule and model contribution back to its version. A BRMS that cannot produce that trace is not finished software.

Higson's approach to this is to run ML models through an ONNX runtime inside decision tables, so each model contribution is a logged input to a deterministic rule. The rule decides what to do with the score; the score itself is auditable end-to-end. This is the architecture pattern I recommend regardless of which BRMS a carrier ends up choosing - pure-AI underwriting deployments without a rules layer have, in my experience, struggled most with the audit-trail requirement.

The honest 2026 outlook - AI agents, MCP, and what "autonomous underwriting" really means

I will state two specific opinions here because they are increasingly contested and worth being clear about.

First, AI agents are not going to take over underwriting in the next five years - but they are going to change how underwriters work. The Model Context Protocol (MCP), introduced by Anthropic in late 2024 and adopted by a growing number of vendors through 2025-2026, lets AI agents read carrier data, propose changes, and execute workflow steps under explicit human authorization. Higson exposes an MCP server with over fifty tools. The realistic 2026 use case is not autonomous binding - it is an AI agent that drafts a rule-change proposal based on loss-ratio drift, surfaces it to the CUO with reasoning, and executes the change through the existing approval workflow once the CUO approves. The governance posture stays the same; the speed of suggestion-to-deployment compresses. I have demoed this pattern to multiple CUO + CAIO joint meetings in the last six months, and it is the use case that actually gets serious follow-up conversations started.

Second, "autonomous underwriting" is a category that exists for narrow product lines - simplified life insurance, low-limit specialty, certain personal-lines direct-to-consumer flows - and is not coming to mainstream P&C commercial within the next decade. The reasons are not technical, they are regulatory and operational: combined-ratio responsibility lives with a human, audit obligations land on a human, and the marginal economics of removing the last 25% of human judgment are usually worse than the marginal cost of keeping it. Vendors who pitch "fully autonomous underwriting" to mid-market commercial CUOs are, in my experience, selling a slide deck.

What does work in 2026: hybrid rules + ML for the deterministic 60-75%, human underwriters concentrated on the complex tail, AI agents augmenting the rule-tuning loop, governance built into the architecture from week one. That is the realistic envelope. The rest is forecasting.

Reference cases - InterRisk, Warta, Notus

Three concrete examples that illustrate the patterns above.

InterRisk (Vienna Insurance Group) - 22 minutes to 4 minutes

InterRisk's Digital Sales Platform Transformation paired a multi-product quote-to-bind experience with BRMS-powered underwriting. Within six weeks of go-live, average quote-to-bind dropped from 22 minutes to 4 minutes. Their CUO mentioned to me, weeks later, that the unexpected benefit was service-center deflection - roughly 80% of agents stopped calling to ask where their quote was. That moment, more than any STP percentage, is when modernization paid for itself.

Warta - 47% fewer manual referrals across 12 lines

Warta consolidated 12 product lines onto a single Higson rules platform, replacing four separate rule-management systems. After six months, manual referral rate dropped by approximately 47% and rule deployment time dropped from quarterly to weekly. Their CUO's anchor quote: "For the first time in 20 years, when an examiner asked how we ensure consistent rule application across states, I had one screen to show them." That is what modern underwriting technology looks like when it lands.

Notus Finance - banking-adjacent underwriting, 100K calcs in 8 seconds

Notus Finance, a top-3 financial intermediary in Poland specialising in mortgages, migrated from Drools to Higson for their broker decisioning. The throughput headline was 100,000 calculations in 8 seconds on the new engine, against roughly 14 seconds previously. The deeper reason this case matters in an underwriting article is that the same engine pattern - rules + reference data + audit trail - generalizes from mortgage broker decisioning to insurance underwriting without changing the architecture. Higson's cross-vertical workload is part of why the product behaves consistently across mid-market P&C and banking-adjacent underwriting.

FAQ - modern underwriting technology

What is modern underwriting technology in plain terms?

Modern underwriting technology is the stack of tools that helps insurers evaluate, price, and bind risk faster and more consistently. The core components are a business rules engine (for deterministic policy execution), embedded ML models (for predictive risk scoring), real-time data enrichment (credit, MVR, telematics, IoT, satellite imagery), an underwriter workbench for referred cases, and an audit layer that satisfies NAIC and state-DOI requirements.

How does modern underwriting improve loss ratio for mid-market carriers?

Three ways. First, by enforcing consistent rule application across underwriters and states, removing the inconsistency that drives mid-market loss ratio drift. Second, by feeding richer real-time data (telematics, IoT, geospatial, document AI extraction) into risk scoring. Third, by freeing senior underwriters to concentrate on the complex tail where their judgment moves outcomes most. McKinsey's research puts the realistic mid-market loss-ratio improvement at 3-5 combined-ratio points, with up to 10-15% lift in business premiums for leading insurers.

What is the difference between a underwriting workbench and a underwriting engine?

A underwriting workbench is the interface where underwriters handle referred cases - the queue, the override workflow, the documentation. A underwriting engine is the back-end decisioning system that determines which cases reach the workbench at all. Carriers buying one and expecting the other is a recurring source of failed modernization budgets. Modern stacks usually pair a workbench from Guidewire, Duck Creek, Insurity, or Sapiens with a specialized underwriting engine (BRMS) like Higson for the decisioning layer.

Will AI replace human underwriters by 2030?

No. AI will replace specific underwriter tasks - data entry, document extraction, simple eligibility checks, first-pass risk scoring - but the underwriter role evolves rather than disappears. Senior underwriters increasingly focus on the complex risk tail, portfolio strategy, and governance over AI/ML deployments. NAIC's audit-trail requirements and combined-ratio accountability both keep human oversight in the loop. Any vendor pitching "AI replaces underwriters" is, in my experience, selling marketing rather than architecture.

What is a realistic STP rate for modern underwriting?

Sixty to seventy-five percent for mid-market personal lines (auto, home) with a properly designed rules engine and clean data feeds. Fifty to sixty-five percent for mid-market commercial multi-line. Direct writers like Progressive and GEICO can run 80%+ on consumer-direct flows. Specialty and E&S realistically reach 20-40% with document AI investment. 100% STP for a non-direct, multi-line book is a marketing claim, not an architecture target.

Is AI underwriting NAIC compliant in 2026?

It can be, provided the carrier maintains a full audit trail of inputs, model versions, contributing features, rule executions, and human overrides - and has a written AI governance program covering data quality, model validation, fairness testing, drift monitoring, and override controls. NAIC's Model Bulletin on AI Use in Insurance (2023, updated through 2024-2025) makes both requirements explicit. Colorado SB 21-169 adds an explainability requirement for external consumer data and ML influence. Multi-state carriers should implement to the strictest applicable standard.

How long does it take to deploy modern underwriting technology?

Three to six months for a single line of business or product family in a mid-market carrier, given a clear appetite map, available data feeds, and a co-located underwriting and architecture team. Multi-line consolidations run six to twelve months. Enterprise full PAS replacements at $5B+ GWP run 18-36 months, but that is usually a Guidewire- or Duck-Creek-led program rather than a focused underwriting engine deployment.

How do IoT and telematics fit into modern underwriting?

Telematics - driving behavior, mileage, time-of-day, acceleration patterns - is now a routine input for many personal auto underwriting programs, alongside the traditional MVR record. IoT signals from smart home devices (water-leak sensors, smart thermostats, smoke detectors) are increasingly used for homeowners and small-commercial property, sometimes for mid-term re-rating rather than just initial pricing. A BRMS lets the underwriting team decide which signals trigger which rule paths without an engineering ticket - which is the difference between deploying telematics in months and deploying it in quarters.

Related reading and how to talk to Higson

• What is a Rules Engine? Complete Guide - the BRMS foundations under every modern underwriting deployment.

• Business Rules Engine vs Decision Engine: Key Differences - for architects untangling vendor naming.

• Underwriting Automation with Rules Engines: 2026 Architect's Guide - the decision-table mechanism explained for Daniel + CUO.

• Automated Underwriting Systems (AUS) — Insurance and Mortgage Guide - cross-vertical AUS framing.

• Straight-Through Processing in Insurance - STP deep-dive.

• Underwriting Efficiency with Business Rules - manual-referral reduction.

• Business Rules Management System — use case - BRMS use case from a underwriting perspective.

• Decision Tables in Higson - what your underwriting team actually edits.

Talk to Higson

If you are scoping modern underwriting technology for a mid-market carrier - whether you are starting at stage 2 or pushing from stage 3 to stage 4 - the most useful 30 minutes you can spend is a joint working session with your CUO and your VP Product. I am happy to walk through your data sources, your appetite map, your STP target, and your state-scope obligations. I will be specific about where Higson fits cleanly and where another vendor would serve you better.

• Download the Business Rules Engine Comparison Guide (9-criteria buying checklist, PDF lead magnet)

• Book a 30-minute modern underwriting working session

• Try Higson on AWS Marketplace at $0.63 per hour. Full BRMS runtime, no procurement cycle