Submission intake
AI reads broker emails, portal uploads and attachments, classifies the line of business and appetite, and flags duplicates and clearance conflicts.
AI in insurance underwriting: use cases, data and regulation
Underwriters still spend more than a third of their time on non-core work, fewer than half trust automated recommendations, and regulators in 24 states and DC now expect a written AI program. Here is what the data says, and where AI helps underwriters decide faster without taking the decision away from them.
- Underwriters still spend more than a third of their time on non-core work (Accenture, 2024 survey)
- 43% of underwriters trust and regularly accept automated recommendations (Capgemini, 2024)
- 24 states and DC had adopted the NAIC Model Bulletin on insurers' use of AI as of April 2026
Insurance underwriting and AI in numbers
- 65%is Accenture's upper estimate for underwriting working hours that AI could automate or augment
- 43%of underwriters trust and regularly accept recommendations from predictive analytics tools (2024)
- 24US states, plus DC, had adopted the NAIC Model Bulletin on insurers' use of AI as of April 1, 2026
- 39.2%of US finance and insurance businesses used AI in the two weeks to October 18, 2026
Where underwriting time goes
Accenture's 2024 underwriting survey found that more than a third of an underwriter's time still goes to non-core work such as collecting data and administration.
The tools are part of the problem. Senior underwriting executives surveyed for Accenture's 2025 report named ineffective systems as their top challenge (65%), ahead of missing analytics (42%) and poorly organized information (40%).
Executives expect AI to reach the decision itself over time. In our experience, the fastest and lowest-risk gains come first from the work around it: reading submissions, extracting data, chasing missing information and preparing the case for the underwriter.
What underwriting teams struggle with
Capgemini's 2024 research across 18 markets shows that the problem is not effort but information: most insurers cannot evaluate risk completely or decide consistently.
- Incomplete risk evaluation77%
- Limited pricing accuracy73%
- Inconsistent underwriting decisions70%
Capgemini Research Institute, World Property and Casualty Insurance Report 2024. Top barriers named: data access (54%), legacy systems (51%) and lack of skilled talent (47%).
View data as a table
| Incomplete risk evaluation | 77% |
|---|---|
| Limited pricing accuracy | 73% |
| Inconsistent underwriting decisions | 70% |
AI use cases across the underwriting workflow
AI works best as an assistant at each step of the workflow, with the underwriter keeping the decision and the authority. Pick a stage to see what AI does and which systems it needs.
Document extraction
AI turns ACORD applications, schedules of values, loss runs and financials into structured fields, each with a confidence score and a link to the source page.
Missing information chase
AI checks the submission against underwriting guidelines and drafts the follow-up to the broker for the underwriter to approve.
Risk enrichment
AI attaches property, catastrophe, firmographic and other third-party data, where its use is lawful, to the submission.
Guideline and precedent copilot
Underwriters ask questions of manuals and authority letters and see similar past accounts with their outcomes.
Rating support
AI prefills rating inputs, explains the gap between technical and quoted price and flags deviations from filed rates.
Referrals
Cases above authority limits go to a senior underwriter with a ready case summary and recommendation. A person decides.
Decision audit trail
Every recommendation is logged with its inputs, model version and reviewer, which gives you the documentation regulators can request under the NAIC bulletin.
Portfolio monitoring
Drift and bias checks run on live decisions, and credit underwriting gets the specific adverse-action reasons Regulation B requires.
Underwriting is already a top AI function
When the NAIC surveyed 194 large US home insurers in 2022–23, underwriting tied with marketing as the second most common place they used advanced AI and machine learning models, after claims.
- Claims54%
- Underwriting47%
- Marketing47%
- Fraud detection42%
- Rating35%
- Loss prevention14%
NAIC Big Data and AI Working Group home insurer survey, analysis dated August 2023. Only more advanced AI/ML models were counted.
View data as a table
| Claims | 54% |
|---|---|
| Underwriting | 47% |
| Marketing | 47% |
| Fraud detection | 42% |
| Rating | 35% |
| Loss prevention | 14% |
Not sure where AI fits in your underwriting workflow?
We review your intake, policy admin and rating systems and show which step saves underwriters the most time.
Knowledge transfer is a real need
About 24% of employed US insurance underwriters are 55 or older, close to the 23% across all US workers. As experienced underwriters retire, 76% of underwriting executives surveyed by Accenture expect AI to help pass their knowledge to new underwriters.
- 20–242K
- 25–3440K
- 35–4432K
- 45–5433K
- 55–6423K
- 65 and over11K
Current Population Survey 2025 averages, which exclude October. The BLS projects underwriter employment to fall 4% from 2025 to 2035, yet still expects about 6,800 openings a year.
View data as a table
| 20–24 | 2K |
|---|---|
| 25–34 | 40K |
| 35–44 | 32K |
| 45–54 | 33K |
| 55–64 | 23K |
| 65 and over | 11K |
Where the data disagrees: automate decisions or assist underwriters?
Executives plan for heavy automation. Underwriters, regulators and production data suggest most AI will assist rather than decide for some time.
AI adoption is growing fast
- Underwriting executives expect their use of AI technologies to grow from 14% today to 70% within three years.
- In a 2023 NAIC survey of life insurers, 48% of pricing and underwriting models, including generalized linear models, automate the decision.
- Almost 65% of EU insurers already use generative AI somewhere in the business.
Assistance is what works today
- Only 43% of underwriters trust and regularly accept automated recommendations, citing complexity and data integrity.
- 64% of generative AI use cases at EU insurers are still proofs of concept, and only 32% are in production.
- Regulators expect a documented AI program, with senior management accountable for how AI is used.
Evgeny Lupanov
Chief Technical Officer
Both are true, and the order matters. In our experience, insurers that start by automating decisions run into trust and compliance questions. Those that start with intake, extraction and case summaries give underwriters time back and build the data and audit trail that later automation needs.
[ Regulation ]
What regulators expect from AI in underwriting
NAIC Model Bulletin
Adopted in December 2023, it expects insurers to keep a written AI Systems program covering underwriting, rating and pricing, including models bought from vendors. 24 states and DC had adopted it as of April 2026.
State rules on external data
Colorado's SB21-169 prohibits unfair discrimination through external consumer data, algorithms and predictive models. California, Colorado, New York and Texas also have their own insurance AI regulation or guidance.
Credit underwriting needs specific reasons
Under Regulation B, an adverse action notice must state the principal reasons. Saying the applicant failed the creditor's scoring system is not enough.
EU AI Act high-risk rules
Risk assessment and pricing of individuals in life and health insurance, and credit scoring of individuals, are high-risk uses. P&C and commercial lines are not on that list. After the AI Omnibus, these rules apply from 2 December 2027.
“Underwriters do not need a model that replaces their judgment. They need one that reads the submission, finds what is missing and shows where every number came from. Build that first, keep the underwriter as the decision maker and log everything. Automation of simple risks comes later, on top of data you can defend to a regulator.”
[ Case Studies ]
Insurance work we have delivered
Neither is an underwriting project, but both involve the document handling, workflows and integrations an underwriting assistant needs.
InsuranceAI Development
AI Platform for Insurance Brokers
1,500+ broker hours saved a year
For an AI insurtech that automates quote comparison and proposals, we redesigned more than 10 core broker workflows around its AI engine and integrated payments, so brokers spend less time in the interface.
InsuranceWeb Development
Insurance Technology Platform for Accuro Solutions
2 production platforms delivered
We built the ARC Review frontend, which automates medical bill review for casualty insurers, and the full ARC Finance payment system, delivered in 12 and 6 months.
Sources
- Accenture, AI underwriting beyond the hype · 2024 underwriting survey, published September 2025
- Accenture, Underwriting rewritten · 430 senior underwriting executives in 11 countries, August 2025
- Capgemini Research Institute, World Property and Casualty Insurance Report 2024 · Underwriter, executive and customer surveys in 18 markets, April 2024
- NAIC, Implementation of the Model Bulletin on the Use of AI Systems by Insurers · Status as of April 1, 2026
- NAIC, Home AI/ML Survey Analysis · 194 home insurers, August 2023
- NAIC, Life Insurance AI/ML Survey Results · 161 life insurers, December 2023
- US Bureau of Labor Statistics, Occupational Outlook Handbook: Insurance Underwriters · 2025–2035 projections
- US Bureau of Labor Statistics, Employed persons by detailed occupation and age · Current Population Survey, 2025 annual averages
- EIOPA, Generative AI market survey · 347 insurers in 25 EU/EEA countries, February 2026
- US Census Bureau, Business Trends and Outlook Survey · Finance and insurance sector, October 5–18, 2026
- Colorado General Assembly, SB21-169 · Signed July 2021
- European Commission, AI Omnibus enters into force · July 2026
[ FAQ ]
Frequently asked questions
Discuss AI for your underwriting team
Tell us about your lines of business, submission volumes and the systems your underwriters use. We will come back with the step where AI pays off fastest and a rough estimate.
Dmitry Klim
Head of Growth
