Underwriting Analysis with AI

Underwriting is the part of insurance that decides which risks the carrier accepts, on what terms, at what price. It is heavy on judgment and heavy on document review. AI is not going to replace underwriting judgment — but it can do a lot of the document and analysis work that underwriters do every day.

What You'll Learn

• How to use AI for submission analysis and risk profiling
• Prompts for SIC/NAICS classification, industry research, and exposure analysis
• How to extract data from ACORD applications and loss runs
• The ethical limits of AI-driven underwriting

What Underwriters Actually Do

A typical workday for a P&C or specialty underwriter includes:

• Reviewing submission packets (ACORD applications, supplemental questionnaires, loss runs, financials)
• Researching the prospect's industry and operations
• Identifying coverage exposures and recommending limits
• Pricing the risk using the rating engine
• Negotiating terms with the broker
• Documenting the underwriting decision

Of these, the parts most amenable to AI are review, research, and documentation. Pricing belongs in the rating engine. Decisions belong with the underwriter.

Submission Triage

Many lines see a high volume of submissions, only a fraction of which fit appetite. AI can do a first-pass triage in seconds.

You are a small commercial lines underwriter for a US carrier.
Our appetite for BOP includes professional services,
restaurants under $2M revenue, retail under $3M revenue, and
light contractors. We avoid: cannabis, firearms, adult
entertainment, mold remediation, and demolition. Maximum
property TIV $5M per location.

Below is a submission summary. Tell me:
1. Does this fit appetite? (Yes / Conditional / No)
2. The single biggest reason for the answer
3. Three follow-up questions for the broker if conditional
4. If no, the top alternative market category to suggest

Submission summary:
[paste 1-paragraph description from the broker email]

This kind of prompt lets a single underwriter triage 100 submissions in the time it used to take to triage 30.

Reading ACORD Applications

ACORD applications are structured but lengthy. AI can extract the fields that matter most for your appetite analysis.

You are a commercial lines underwriter. From the ACORD 125
text below, extract:

- Applicant legal name and DBA (if any)
- Years in business
- Number of locations
- Total annual revenue
- Total payroll
- Number of employees
- SIC and NAICS codes
- Operations description (verbatim)
- Prior carrier information (if listed)
- Any Yes answers to general information questions
  (with brief context)

If a field is missing or illegible, mark it "Not provided".
Do not infer values not present.

ACORD 125 text: [paste]

For supplemental applications (restaurant, contractor, professional liability), build a similar prompt template per supplemental.

Loss Run Analysis

Loss runs are where the story of a risk lives. AI can summarize and pattern-match across multiple years quickly.

You are a commercial lines underwriter analyzing a 5-year loss
run for a medium-sized risk. Below are the loss entries.
Produce:

1. Total incurred by year
2. Frequency by line of coverage
3. Three patterns or trends worth flagging
4. Three loss types that drove the largest dollar amount
5. Any losses with open reserves > $50,000
6. Three follow-up questions for the broker

Constraint: Do not invent any loss data. If years or columns
are missing, note that explicitly.

Loss run text: [paste]

Industry and Operations Research

Underwriting often requires understanding what an applicant actually does. Perplexity is particularly useful here because it cites sources.

A useful Perplexity prompt:

I am underwriting a commercial general liability submission
for [INDUSTRY] in [STATE]. Summarize:

- The 3 most common claim types in this industry
- 2 emerging risk trends in 2025-2026
- Typical safety standards or industry certifications
- 3 underwriting questions specific to this industry

Cite sources I can verify.

For deeper research on a specific applicant (publicly traded or with significant web presence), you can ask Perplexity or ChatGPT with web access to summarize public filings, news mentions, and known incidents.

Risk Profiling

Once you have a submission, you want a structured risk profile.

You are a senior P&C underwriter. Build a structured risk
profile from the information below. Output:

- Risk type and operations summary
- Top 5 exposures by line of coverage
- 3 strengths of this risk
- 3 concerns about this risk
- Recommended coverage forms (general categories, not
  specific carrier forms)
- Recommended limits range
- 3 underwriting actions to take before binding

Constraint: Do not recommend specific carrier products.
Do not commit to specific premium. Use the information
provided only.

Information: [paste application + loss summary]

Underwriting Memos

Many carriers require underwriting memos for accounts above a certain size. AI can produce a clean first draft from your notes.

You are a senior commercial underwriter writing an
underwriting memo for an account over $250,000 in premium.
From my notes below, draft a memo with sections:

- Account overview
- Coverage and limits offered
- Pricing rationale (qualitative — refer to "rating
  worksheet" rather than inventing numbers)
- Exposures and mitigating factors
- Prior loss experience (summarize what is in the notes)
- Recommended action

Notes: [paste]

The Ethical Line

There are clear places where AI should not be used in underwriting decisions:

• Protected class proxies. Do not let AI use ZIP code, name, or other variables that proxy for race, ethnicity, religion, or national origin in declination decisions.
• Black-box pricing. Pricing must be explainable. Most state DOIs are tightening rules on AI in rating.
• Adverse-action automation. Declinations and non-renewals trigger consumer notice rights under the Fair Credit Reporting Act and state law. Always document the human decision.
• Personal lines AI scoring without disclosure. Several states now require disclosure when AI plays a role in personal lines underwriting decisions.

Use AI as a research and drafting assistant, not as the decision-maker.

A Note on the NAIC AI Bulletin

The NAIC adopted a Model Bulletin on the Use of AI Systems by Insurers in late 2023, and a growing number of states have implemented it. The bulletin asks carriers to maintain an AI governance program, document AI use cases, monitor for bias, and ensure consumer protections. If you are an underwriter at a regulated carrier, your AI use should fit within whatever program your carrier has established.

Key Takeaways

• AI can dramatically accelerate the parts of underwriting that involve document review, extraction, research, and drafting. It does not replace underwriting judgment or rating.
• Submission triage prompts let you sort fits, conditionals, and declines at high volume.
• ACORD and loss run extraction prompts produce clean structured data from messy source documents.
• Underwriting decisions, especially declinations and non-renewals, must be made and documented by a licensed underwriter — not by AI.