AI in Insurance (2026): Use Cases in Underwriting, Claims and Distribution

The short answer: insurers use AI across the value chain: reading broker submissions and triaging them for underwriters, building pricing and risk models, taking first notice of loss and routing claims, estimating damage from photos, detecting fraud, summarising claim files and medical records, and helping agents and service teams answer coverage questions. The strongest cases remove manual reading and data entry while a licensed underwriter, adjuster or investigator keeps the decision. Insurance is also one of the most regulated places to use AI, so governance, documentation and fairness testing must be part of the plan from the start.

This guide is for carrier, MGA, broker and agency leaders in operations, underwriting, claims and IT. Tool names are examples of each category, not endorsements. Most insurance AI is quote-based, priced per policy, per claim, per user or in relation to premium volume, so check each vendor’s current pricing directly.

AI use cases in insurance at a glance

Use case	What the AI does	Tool examples	Effort	Payback signal
Submission intake and triage	Extracts data from broker emails, applications and schedules; scores fit with appetite	Cytora, document AI platforms	Medium	Time to quote, submissions handled per underwriter, hit ratio
Risk assessment and pricing models	Predicts loss likelihood and severity from policy, claims and third-party data	Pricing platforms such as hyperexponential, in-house actuarial models	High: data, actuarial and regulatory review	Loss ratio by segment, pricing turnaround
FNOL and claims triage	Captures first notice of loss by chat, phone or app; scores complexity and routes the claim	Core claims systems (Guidewire, Duck Creek), claims messaging such as Hi Marley	Medium	Cycle time, touches per claim, customer satisfaction
Photo-based damage estimating	Assesses vehicle or property damage from photos and drafts an estimate	Tractable, CCC Intelligent Solutions	Medium	Days to estimate, reinspection rate
Fraud detection	Scores claims and policies for fraud indicators and links related parties	Shift Technology, fraud analytics platforms	Medium	Referrals accepted by SIU, confirmed fraud savings
Claim file and medical record summaries	Summarises long files, medical records and correspondence for adjusters	Document AI, claims platform AI features	Low to medium	Adjuster hours per file, reserve accuracy
Agent and service assistants	Answers coverage and billing questions with citations to the policy; drafts emails	Agency management system AI features, AI support agents	Low to medium	Handle time, first-contact resolution
Subrogation detection	Flags claims where another party may be liable	Claims analytics tools	Medium	Recoveries per 1,000 claims

How is AI used in underwriting?

Submission intake and triage

The job: commercial underwriters receive far more submissions than they can quote. The work is sorting, re-keying and chasing missing information.

• Input: broker emails with attachments: applications, ACORD forms, statements of values, loss runs and schedules.
• AI step: document AI classifies each attachment, extracts fields into structured data, enriches it with third-party data (business details, property characteristics), checks it against underwriting appetite and scores priority.
• Human review: an underwriting assistant checks low-confidence fields; the underwriter decides whether to quote and on what terms.
• Output: a clean, prioritised submission queue in the underwriting workbench or policy system.

What it needs: a clear appetite definition, access to the submission mailbox and integration with the policy administration system. Cytora is one example of a platform built for this. Measure it by: time from submission to quote, submissions handled per underwriter and hit ratio on prioritised business. Risks: extraction errors on values and loss history that flow into pricing. Keep field-level confidence scores and review rules.

Risk assessment and pricing

Machine learning models estimate the probability and severity of loss from policy, claims and external data, such as aerial imagery of roofs or telematics for auto. Actuaries validate the models, and pricing changes go through the normal rate filing process where required. Pricing platforms such as hyperexponential help actuarial teams build and deploy rating models faster. This is the area regulators watch most closely. Colorado’s law on insurers’ use of external consumer data and algorithms (SB21-169) requires testing for unfair discrimination, and the EU AI Act treats AI used for risk assessment and pricing in life and health insurance as high risk. Measure loss ratio by segment against a holdout and time to deploy rate changes. Document every model: data sources, variables, testing and approvals.

How is AI used in claims?

First notice of loss and claims triage

The job: capture the claim quickly and accurately, and send it to the right handler, so simple claims close fast and complex ones get experienced adjusters early.

• Input: the policyholder’s report by app, web chat, text or phone, plus photos and policy data.
• AI step: a conversational assistant collects the facts, checks coverage basics against the policy, scores complexity, severity and fraud indicators, and suggests a route: straight-through, desk adjuster, field adjuster or special investigations.
• Human review: a claims supervisor sets routing rules and reviews exceptions; coverage decisions stay with licensed adjusters.
• Output: a complete claim file, assigned to the right person on day one.

Core claims systems such as Guidewire and Duck Creek support this kind of workflow, and claims messaging tools such as Hi Marley handle text conversations between adjusters and policyholders. Measure it by: claim cycle time, touches per claim, reassignment rate and customer satisfaction. Risks: policyholders who cannot reach a person during a stressful loss; always offer an easy route to a human.

Damage assessment from photos

For auto claims, computer vision reviews photos of the vehicle, identifies damaged parts, estimates repair or replace decisions and drafts an estimate; Tractable and CCC Intelligent Solutions are established examples. For property, AI helps sort photos and scope damage, while estimators still build line-item estimates in tools such as Xactimate (see our Xactimate overview). An appraiser reviews the AI estimate before payment. Measure days from FNOL to estimate and reinspection or supplement rates. The risk is hidden damage that photos cannot show, so set rules for when a physical inspection is required.

Fraud detection

Fraud models score each claim against known patterns, such as recent policy changes before a loss, repeated parties across claims, or inconsistent dates, and network analysis links people, vehicles, addresses and providers across claims. Shift Technology is a well-known specialist. Special investigations unit (SIU) staff decide whether to investigate; the model only refers. Measure the share of referrals the SIU accepts, confirmed fraud savings and false positives that delayed honest claims. See fraud protection software.

Claim file and medical record summaries

Bodily injury, workers’ compensation and disability claims can involve thousands of pages. Generative AI summarises medical records, treatment timelines and correspondence, with links back to the source page, so adjusters spend time on judgement instead of reading. The adjuster verifies key facts before setting reserves or making offers. Measure adjuster hours per file and reserve changes late in the claim. Health data requires HIPAA-grade controls where it applies, including a business associate agreement with the vendor. See OCR software for document capture.

Subrogation

Models read claim notes and data to flag claims where another party may be responsible, such as a rear-end collision or a faulty appliance, so recovery teams see them early. Recovery specialists decide what to pursue. Measure recoveries per 1,000 claims.

How do agencies and service teams use AI?

Agencies and service centres use AI to answer “am I covered for this?” and billing questions with citations to the actual policy, to draft renewal and follow-up emails, to summarise calls, and to prepare cross-sell suggestions for account managers. Licensed staff give coverage advice; the assistant prepares it. Measure handle time, first-contact resolution and renewal retention. Agency management platforms are adding AI features; compare options in insurance agency management software and our guide to the best CRM for insurance agents. For call analysis, see speech analytics software and conversation intelligence software.

Is it true that AI is denying healthcare claims?

This is a live issue. Lawsuits filed against several large US health insurers since 2023 allege that algorithms were used to deny or cut short coverage for care; the insurers dispute the allegations. Regulators have responded: California, for example, passed a law in 2024 requiring that medical necessity decisions in utilisation review be made by a licensed physician or qualified health professional, not by an algorithm alone. For any insurer, the safe design is the same: AI can prepare and prioritise, but a qualified person makes and signs every adverse decision, and the reasoning is documented.

AI governance requirements for insurers

• A written AI programme: the NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted by many US states, expects insurers to maintain a documented AI systems programme covering governance, risk management and oversight of third-party vendors.
• Fairness testing: test models for unfair discrimination, especially where state law requires it, such as Colorado.
• Explainability and adverse action: be able to explain underwriting and claims decisions to regulators and customers, and meet notice requirements such as those under the Fair Credit Reporting Act where credit-based data is used.
• Data use and training: ask vendors whether your policyholder and claims data trains models used for other customers, and get the answer in the contract.
• Privacy: GLBA and state privacy laws in the US, GDPR in Europe, and HIPAA where health data is involved.
• Security and access: SOC 2 Type II or ISO 27001 as the vendor states them, SSO, role-based access and full audit logs of AI recommendations and human decisions.
• Data residency: required in some jurisdictions and by some reinsurance or group policies.

For tools that help manage this, see AI governance software.

Where should an insurer start with AI?

1. Pick a document-heavy, advisory use case: submission intake, claim file summaries or service assistants. AI prepares, people decide, and regulatory exposure is lower.
2. Stand up governance at the same time: an inventory of AI systems, an owner for each, and a review process before production.
3. Baseline the metric: time to quote, cycle time or handle time over the last two quarters.
4. Pilot with one line of business and one team for 90 days.
5. Move to decision-support models (triage, fraud scoring, pricing) only with documentation, fairness testing and human sign-off in place.

How to measure AI ROI in insurance

Split value into expense and loss effects. Expense: hours saved per submission or claim x volume x loaded cost. Loss: improvement in loss ratio, fraud savings and recoveries, measured against a holdout group, because loss results take time to mature. Subtract software, integration, model governance and people costs. Report expense gains early and loss gains only when the data is credible, which for long-tail lines can take years.

Risks and failure modes

• Unfair discrimination through proxy variables in pricing, underwriting or claims models.
• Automated denials without meaningful human review.
• Extraction errors in submissions or claim files that flow into pricing and reserves.
• Vendor opacity: third-party models you cannot explain to a regulator.
• Customer trust: policyholders who feel a machine decided their claim.
• Legacy systems that make integration slow and expensive.

Will insurance jobs be replaced by AI?

AI is changing insurance jobs more than removing them. Underwriting assistants and claims processors spend less time on data entry; underwriters and adjusters get more capacity for complex risks and claims; new roles grow in data science, model governance and AI operations. The industry has long discussed an ageing workforce and talent gaps in underwriting and claims, and AI is one way carriers are responding. Regulation also keeps licensed people accountable for key decisions.

How to choose an AI tool for insurance

• Match the tool to your line of business: personal auto, commercial property, specialty and health have very different needs.
• Ask for references from carriers or MGAs of similar size and lines.
• Check integration with your policy administration, claims and agency management systems.
• Review the vendor’s model documentation and support for your NAIC-aligned governance.
• Get data-use, training and audit terms in writing.

More on enterprise AI: for other industries, see our guides to AI in manufacturing, retail, real estate, construction and banking. For the cross-functional view, read enterprise AI use cases by function and how to implement AI in business. Related guides: intelligent document processing, AI customer service, AI voice agents and AI governance tools.

Frequently asked questions