Industry Guide: Insurance
How to Use Claude in Insurance: A Guide for Underwriting, Claims, and Broking Teams
Read the whole submission, then make the call yourself.
This guide is for underwriters, claims handlers, brokers, and operations teams who want to put Anthropic's Claude to work on submissions, loss runs, claims files, policy wording, and customer letters. It covers what Anthropic and its partners have announced for insurers and what named insurers and brokers have said in Anthropic's own posts.
Learning how to use Claude in insurance starts with the documents that slow every desk: summarizing underwriting submissions and loss runs, comparing policy wordings, drafting FNOL and claims file summaries, checking claims against coverage terms, and writing clear letters to policyholders. Claude reads and drafts fast, and it explains its reasoning in plain language a reviewer can audit. Underwriters and adjusters keep the decisions on risk, coverage, and payment, inside the insurer's regulatory and fairness obligations.
Published
Key Takeaways
- The strongest insurance uses are submission and loss run summaries, policy wording comparison, claims file summaries, and policyholder letters.
- Anthropic's May 2026 financial services release added a Verisk connector for property, casualty, and specialty insurance data.
- Anthropic has published insurance examples from AIG, Newfront, and DXC Technology, plus an unnamed insurer whose underwriting cycle went from weeks to days.
- Coverage, pricing, and payment decisions stay with licensed people, and AI-assisted underwriting must meet fairness rules.
- Policyholder data belongs only in an approved workspace.
Copy and Paste
Prompts You Can Use Today
Underwriting submission summary
You are a commercial property underwriter. Summarize the submission below for a referral decision: insured and operations, locations and construction, values, protection and exposures, prior losses from the loss runs, requested coverage and limits, and anything missing from our submission requirements listed below. End with the questions to send back to the broker. Quote the document for every figure. Submission and requirements: [paste].
Paste your own submission requirements so gaps are judged against your appetite, not a generic list.
Loss run analysis
Here are five years of loss runs for a commercial auto account. Build a table by policy year with claim count, paid, reserved, and incurred, and list the largest claims with cause and status. Then describe the trends in frequency and severity in plain language and flag any open claim with a large reserve change. Show every calculation and do not estimate anything that is not in the data. Loss runs: [paste].
Asking for calculations lets you verify the totals before they reach a pricing discussion.
Policy wording comparison
Compare the two policy wordings below clause by clause for coverage differences that matter to the insured: insuring agreement, definitions, exclusions, conditions, and endorsements. For each difference, quote both wordings side by side and explain in one sentence which is broader and why. Do not give a legal opinion; flag clauses that need review by coverage counsel. Wordings: [paste A] [paste B].
Side-by-side quotes make the comparison checkable in minutes.
Claim file summary against coverage
Summarize the claim file below for the adjuster: date and description of loss as reported at first notice, parties, documents received, what is still outstanding, and the policy provisions that appear relevant, quoting each one. List the questions the adjuster should resolve before a coverage position. Do not recommend accepting or denying the claim. File and policy: [paste].
The coverage decision is the adjuster's. The summary exists to make that decision faster and better documented.
Plain-language policyholder letter
Rewrite the claim decision letter below in plain language for a policyholder with no insurance background. Keep every substantive point, reference, and date exactly as written, explain any technical term in a short phrase, and keep a calm, respectful tone. Add a short list of next steps and who to contact. Do not change the decision or add any promise. Letter: [paste].
Asking Claude to keep every date and reference unchanged protects the letter's legal content.
From Anthropic
What Anthropic offers for Insurance
Agents for financial services
Anthropic's May 2026 release of ten financial services agent templates added a Verisk connector described as providing property, casualty, and specialty insurance data for underwriting, claims, and risk analysis, alongside Excel, PowerPoint, and Word add-ins.
· Source
Anthropic and DXC Technology alliance
A multi-year global alliance announced in June 2026 under which DXC integrates Claude into systems for clients including insurers and banks, and uses Claude to modernize core insurance systems with agentic solutions.
· Source
Anthropic and TCS partnership
A June 2026 partnership to build Claude-powered offerings including claims processing for insurers. Anthropic says Diligenta, TCS's UK life and pensions business, will use Claude to improve service for its policyholders.
· Source
Anthropic and PwC expanded partnership
A May 2026 expansion covering regulated industries, including insurance, in which Anthropic reports an unnamed insurance client compressing underwriting cycles from weeks to days.
· Source
In Practice
How Insurance Teams Use Claude
According to Anthropic's published customer story
AIG
In Anthropic's July 2025 financial services announcement, AIG's chief executive said early rollouts compressed the timeline to review business by more than five times while raising data accuracy.
According to Anthropic's published customer story
Travelers
Anthropic's May 2026 financial services post quotes Travelers' executive vice president and CTO on significantly higher engineering excellence.
According to Anthropic's published customer story
Newfront
According to Anthropic's customer story, the insurance brokerage Newfront uses Claude for a benefits chatbot and for loss run document processing, which now costs substantially less with higher accuracy.
According to Anthropic's published customer story
DXC Technology
Anthropic's customer story on DXC Technology describes Claude classifying insurance documents in under a second and reducing the share of cases that need human judgment.
Illustrative walkthrough
A commercial submission ready for referral
- An underwriter loads the broker's submission, schedules, and loss runs into the approved workspace with the team's submission requirements.
- Claude summarizes operations, values, exposures, and prior losses, quoting the source for every figure.
- Claude lists missing items and drafts questions for the broker.
- The underwriter checks the loss run table against the originals and corrects one policy year.
- The underwriter decides whether to quote, refer, or decline, and records the reasoning.
Insurance and Claude: the Published Numbers
| Figure | Context | Source |
|---|---|---|
| more than 5x | AIG's reported compression of its business review timeline in early rollouts, per Anthropic's July 2025 announcement | Source |
| from 75% to over 90% | AIG's reported improvement in data accuracy, per Anthropic's July 2025 announcement | Source |
| from ten weeks to ten days | Underwriting cycle of an unnamed insurance client, per Anthropic's PwC partnership announcement | Source |
| costs 60% less | Newfront's loss run document processing, per Anthropic's customer story | Source |
Core Workflows
Claude Insurance Use Cases Across the Policy Life Cycle
These are the tasks where AI for insurance returns the most time to underwriters, adjusters, and brokers.
AI for underwriting
Submission summaries, loss run tables, exposure notes, and referral memos drafted from broker documents, with gaps listed for follow-up.
AI claims processing
FNOL summaries, claim file digests, coverage provision extracts, and reserve change notes prepared for the adjuster.
Policy wording review
Clause-by-clause comparisons of wordings and endorsements, with quotes side by side and flags for coverage counsel.
AI fraud detection in insurance
Claude does not replace fraud scoring models, but it helps investigators summarize files, spot inconsistencies in statements, and document referrals.
AI for insurance agents and brokers
Coverage summaries for clients, renewal comparisons, benefits questions, and proposal drafts, reviewed by the licensed professional.
Policyholder communication
Plain-language letters, claim status updates, and complaint responses that keep every legal point intact.
Underwriting and Claims
How Insurance Companies Use AI for Underwriting and Claims
Underwriting and claims are both reading jobs before they are judgment jobs. An underwriter receives submissions, schedules, loss runs, and inspection reports in every format. An adjuster receives notices, photos, statements, invoices, and correspondence. Claude reads all of it and produces the structured summary the professional needs to decide, with each figure tied to its source document.
Anthropic's published examples show where this lands. In Anthropic's July 2025 financial services announcement, AIG's chief executive said early rollouts compressed the timeline to review business by more than five times while improving data accuracy. In Anthropic's announcement of its expanded PwC partnership, an unnamed insurance client compressed underwriting cycles from ten weeks to ten days, opening lines of business that had not been economically viable.
Data connections matter here too. Anthropic's May 2026 financial services release added a Verisk connector described as providing property, casualty, and specialty insurance data for underwriting, claims, and risk analysis. Where an insurer has it enabled, Claude can combine that data with the submission in one place.
Agentic Claims and FNOL
FNOL Automation and Agentic AI for Claims
FNOL automation means capturing the first notice of loss, structuring it, and routing it without manual rekeying. Claude can turn a free-text notice, call transcript, or email into a structured record and a first summary for the adjuster. Agentic AI for claims goes further: an agent gathers documents, checks them against the policy, and prepares a file for decision.
Can AI approve insurance claims automatically? Technically some simple claims can be settled straight through by rules, but a general assistant should not make coverage or payment decisions on its own. The better design is an agent that prepares the file and a person who approves it, with every step logged. DXC Technology's customer story on Anthropic's site describes classifying insurance documents in under a second and reducing the share of cases needing human judgment, not removing that judgment.
Generative AI in insurance is also reshaping core systems work. Anthropic has announced alliances with DXC to modernize core insurance systems and with TCS to build Claude-powered claims processing for insurers.
Choosing Tools
Claude vs ChatGPT for Insurance
The Claude vs ChatGPT for insurance question should be tested on closed files from your own book. Take a submission you already underwrote and a claim you already closed, run the same prompts in each tool, and compare accuracy, handling of missing information, and how well each quotes its sources.
In practice, insurers decide on governance: approved deployment, data residency, retention, audit logging, and fit with model risk and fairness testing. Those matter more than small differences in output quality, and your risk and compliance teams will set the approved path.
Whatever tool you use, the habits on this page carry over: paste your own requirements, ask for quotes, keep decisions with licensed staff, and check every figure.
Getting Started
A Safe First Month for an Insurance Team
Test on closed files
Run the prompts on submissions and claims that are already decided. Compare the output with what actually happened.
Load your guidelines
Put underwriting guidelines, claims procedures, and letter templates into the project so output follows your rules.
Pick one desk
Start with one line of business and one task, such as loss run summaries, and measure time saved and errors caught.
Document the review
Record who reviewed each AI-assisted output. Regulators and auditors will ask.
Limits
Fairness, Regulation, and Where People Decide
Is AI underwriting fair and regulated? Insurance regulators in many markets now expect insurers to govern AI use, test for unfair discrimination, and explain decisions, and the EU AI Act treats some insurance uses as high risk. Keep Claude on summarizing and drafting, keep rating and eligibility decisions in governed models and with underwriters, and document how every AI-assisted step is reviewed.
Claude can misread a schedule, miss an exclusion buried in an endorsement, or confuse two policy years in a loss run. Every figure that affects price, reserve, or payment must be checked against the source document.
Policyholder data includes health, financial, and personal information protected by law. Use only approved deployments, minimize what you share, and never use personal accounts for claims or underwriting files.
Practice on Real Insurance Documents
Writing precise instructions and checking output are skills you can learn quickly with practice. The free Nightschool AI curriculum is hands-on from the first lesson.
Frequently Asked Questions
How do insurance companies use AI for underwriting and claims?
They use AI to read submissions, loss runs, and claims files, and to produce structured summaries, coverage extracts, and draft letters for the underwriter or adjuster to decide on. Anthropic's posts describe AIG compressing its business review timeline and an unnamed insurer, via PwC, shortening underwriting cycles from weeks to days.
How does AI speed up claims processing?
Mostly by removing reading and rekeying: turning a first notice of loss into a structured record, summarizing documents as they arrive, extracting relevant policy provisions, and drafting letters. The adjuster spends time on the coverage decision instead of on assembling the file.
Can AI approve insurance claims automatically?
Some simple claims can be settled straight through by rules, but a general assistant like Claude should not make coverage or payment decisions by itself. The sound design is an AI that prepares the file and a licensed person who approves it, with every step logged.
Is AI underwriting fair and regulated?
Regulators increasingly expect insurers to govern AI, test for unfair discrimination, and explain decisions, and some insurance uses are treated as high risk under the EU AI Act. Keeping Claude on summarization and drafting, with rating and eligibility decisions made by governed models and underwriters, keeps that risk manageable.
Is Claude used by insurance companies?
Yes. Anthropic's published posts and customer stories include AIG, Travelers, the brokerage Newfront, and DXC Technology's insurance work, and Anthropic has announced insurance-related alliances with DXC and TCS.
What is FNOL automation?
FNOL automation is capturing the first notice of loss from a call, form, or email and turning it into a structured claim record that routes itself without manual rekeying. Claude can structure free-text notices and write the first summary for the adjuster.
Can brokers and agents use Claude?
Yes, for client-facing summaries, renewal comparisons, benefits questions, and proposal drafts. Anthropic's customer story on Newfront describes a benefits chatbot and loss run processing. Anything given to a client should be reviewed by the licensed professional.
Is it safe to put policyholder data into AI?
Only through a deployment your company has approved after data protection and vendor risk review. Policyholder data often includes health and financial information protected by law, so minimize what you share and never use personal accounts.
Keep going