AI Playbook
for Insurance

• Carriers and MGAs
• Agencies and brokers
• Claims organizations and TPAs
• Underwriting, claims, SIU, and actuarial leaders
• Compliance, operations, and finance teams

• Start with: underwriting case summary or loss run summary
• Save as a skill: summary format and guideline references
• Connect: Verisk underwriting connector, submission documents
• First agent: submission triage assistant
• Measure: time to quote, file quality

• Start with: renewal book review or coverage review talking points
• Save as a skill: renewal and prep formats
• Connect: agency management system AI
• First agent: agency renewal book watch
• Measure: retention, producer time saved

# Claim File Summarizer

## Job to be done
Give adjusters an organized file at first review.

## When it runs
When new documents are added to a claim.

## Approved context
Claim file contents, policy provisions.

## Operating loop
Summarize facts; quote provisions; list gaps; flag injury or emergency indicators.

## What it delivers
An updated file summary.

## Human owner
Assigned adjuster.

## What it must never do
Never decide coverage, reserves, or payment.

## How to measure it
Time to first decision, re-contact rate.

## Operating rules
1. Use only approved sources and follow the linked insurance skill.
2. Cite policy, claim, or submission IDs and separate facts from assumptions.
3. Route regulatory deadlines, injury or emergency indicators, and possible unfair treatment to the human owner immediately.
4. Keep an audit record of trigger, inputs, output, reviewer, decision, and override, consistent with your written AI program.
5. Stop and ask for human review when policy language, guidelines, law, or authority are unclear.
6. Never handle personal or health information outside approved systems.

# Subrogation Opportunity Finder

## Job to be done
Catch recovery opportunities before they're lost.

## When it runs
Weekly; at claim closure.

## Approved context
Claim details, causes, third parties, payments.

## Operating loop
Flag claims with recovery potential; list evidence needed.

## What it delivers
A subrogation referral list.

## Human owner
Subrogation manager.

## What it must never do
Never send demands or contact third parties.

## How to measure it
Recoveries, referral rate.

## Operating rules
1. Use only approved sources and follow the linked insurance skill.
2. Cite policy, claim, or submission IDs and separate facts from assumptions.
3. Route regulatory deadlines, injury or emergency indicators, and possible unfair treatment to the human owner immediately.
4. Keep an audit record of trigger, inputs, output, reviewer, decision, and override, consistent with your written AI program.
5. Stop and ask for human review when policy language, guidelines, law, or authority are unclear.
6. Never handle personal or health information outside approved systems.

# Notice Compliance Checker

## Job to be done
Stop defective cancellation and non-renewal notices before mailing.

## When it runs
Before each notice batch.

## Approved context
Notice queue, state requirements.

## Operating loop
Check advance notice and reason requirements; flag defects.

## What it delivers
A notice exception list.

## Human owner
Policy service supervisor.

## What it must never do
Never mail or cancel.

## How to measure it
Defective notices caught.

## Operating rules
1. Use only approved sources and follow the linked insurance skill.
2. Cite policy, claim, or submission IDs and separate facts from assumptions.
3. Route regulatory deadlines, injury or emergency indicators, and possible unfair treatment to the human owner immediately.
4. Keep an audit record of trigger, inputs, output, reviewer, decision, and override, consistent with your written AI program.
5. Stop and ask for human review when policy language, guidelines, law, or authority are unclear.
6. Never handle personal or health information outside approved systems.

# Monthly Results Narrator

## Job to be done
Cut the time to produce monthly results narratives.

## When it runs
Monthly after close.

## Approved context
Results by line, plan, prior year, cat losses.

## Operating loop
Compute ratios; explain drivers; list questions.

## What it delivers
A draft narrative.

## Human owner
Controller or CFO.

## What it must never do
Never change reserves or distribute reports.

## How to measure it
Prep time, edits needed.

## Operating rules
1. Use only approved sources and follow the linked insurance skill.
2. Cite policy, claim, or submission IDs and separate facts from assumptions.
3. Route regulatory deadlines, injury or emergency indicators, and possible unfair treatment to the human owner immediately.
4. Keep an audit record of trigger, inputs, output, reviewer, decision, and override, consistent with your written AI program.
5. Stop and ask for human review when policy language, guidelines, law, or authority are unclear.
6. Never handle personal or health information outside approved systems.

• What AI does: Extracts data from applications, loss runs, inspections, and supplements
• Structures: Documents into underwriting inputs
• Control: Underwriter verifies key fields

• What AI does: Assembles forms and endorsements per rules
• Delivers: Accurate documents faster
• Control: Underwriting reviews before issue

• What AI does: Captures claim details from voice, chat, or forms
• Creates: Structured intake summaries
• Control: Injury and emergency flags go to people now

• What AI does: Line-by-line estimate differences
• Supports: Appraisers and adjusters
• Control: People set payment amounts

• What AI does: Ranks leads by fit and readiness
• Prioritizes: Producer time
• Control: Test for unfair treatment

• What AI does: Flags large increases and remarketing needs
• Improves: Retention
• Control: Producer decides

• What AI does: Scores claims for referral review at intake
• Flags: Files meeting written criteria
• Control: Scores are leads, not findings

• What AI does: Flags outlier billing for review
• Supports: Medical fraud investigations
• Control: Provider rights apply

• What AI does: Checks requests against referral rules
• Speeds: Service turnaround
• Control: Underwriting referrals respected

• What AI does: Payments, lapses, and reinstatements
• Prevents: Unintended lapses
• Control: Supervisor decides

• What AI does: Every AI system, use, owner, and risk tier
• Helps: Exam readiness
• Control: Governance committee confirms

• What AI does: Unfair claims practices timeframes
• Helps: Fewer exam findings
• Control: Supervisors act

• Ask AI to do one clear task
• Use it for a quick draft, analysis, or explanation
• Check the result before you use it

• Run a skill on a schedule or when something happens
• Gather the information, prepare the work, and flag issues
• Leave underwriting, claims, fraud, and advice decisions with licensed people

• Faster submission triage and summaries
• More time on judgment, less on reading
• Consistent use of guidelines
• Better documentation for audits

• Consistent referral criteria
• Link analysis summaries
• Better investigator focus
• Fairness monitoring of alerts

Interprets policy documents, claim narratives, and applications to extract meaning.

In insurance: policy analysis, claim narrative review, guideline Q&A.

Flags unusual patterns compared to normal.

In insurance: unusual claim patterns, pricing outliers, billing anomalies.

Reads images and video.

In insurance: damage photo review and property imagery, verified by people.

• Test underwriting, pricing, claims, and fraud models for unfair discrimination
• Review external data sources before use
• Document tests and remediation
• Follow state rules (e.g., Colorado, New York)

• Underwriters decide acceptance and terms
• Adjusters decide coverage and payment
• Investigators decide fraud findings
• Producers give advice