Automating Claims Triage with AI
Claims triage is the process of receiving an FNOL and routing it to the right team, with the right severity, with the right initial reserves. Most carriers do this with rules engines today; AI is rapidly augmenting those rules with better text understanding. Even without a sophisticated platform, you can use AI to dramatically speed up your personal triage workflow.
What You'll Learn
- How to use AI to classify and route incoming claims
- Severity scoring prompts that produce consistent ratings
- Initial reserve recommendation prompts
- How to chain prompts into a workflow
The Triage Decision Stack
Every triage event involves a small stack of decisions:
- Coverage tier: Is this likely covered, likely not, or ambiguous?
- Severity: Is this a small, medium, or large loss?
- Routing: Which adjuster team or specialty handles this?
- Reserve: What is the initial financial reserve?
- Special handling: Does this need SIU, attorney, or executive review?
AI can produce a recommendation on each one. A licensed adjuster reviews and confirms before action.
Coverage Tier Prompt
You are a senior commercial property claims supervisor.
Below is an FNOL and a one-paragraph policy description.
In ONE WORD answer: COVERED, NOT_COVERED, or AMBIGUOUS.
Then in 2-3 sentences explain why.
Constraint: Base the answer only on the policy description
provided. Do not invent policy language.
FNOL: [paste FNOL narrative]
Policy description: [paste short policy description]
This is not a coverage decision; it is a triage signal that helps you prioritize file review.
Severity Scoring Prompt
You are an experienced claims supervisor. Based on the
FNOL below, classify this loss on a severity scale:
1 = Small (estimated total incurred < $10,000)
2 = Medium ($10,000 - $100,000)
3 = Large ($100,000 - $500,000)
4 = Major ($500,000 - $2,000,000)
5 = Catastrophic (> $2,000,000)
Output:
- Severity score (1-5)
- Confidence (Low / Medium / High)
- 3 factors driving the score
- 3 questions that would change the score if answered
FNOL: [paste]
The structure produces consistent severity output across thousands of FNOLs.
Routing Prompt
You are a claims router. Our claim teams are:
- Auto Express (auto property damage only, < $15,000,
no injuries, single vehicle)
- Auto Standard (auto with injuries OR property > $15,000)
- Auto Litigation (auto with attorney representation OR
fatality OR commercial vehicle)
- Property Field (homeowners and small commercial property
losses needing inspection)
- Property Desk (property losses < $5,000 with photos)
- Workers Comp (any WC claim)
- Specialty (cyber, professional liability, environmental)
- SIU Referral (any claim with fraud indicators)
Below is an FNOL. Recommend the team in one line, then
explain in 2 sentences why.
FNOL: [paste]
You can adapt the team list to your carrier's structure.
Initial Reserve Recommendation
You are a claims supervisor. Recommend an initial reserve
for the loss described below.
Output:
- Recommended initial reserve amount
- Reserve breakdown (indemnity, expense, ALAE)
- Three reasons supporting the amount
- Three factors that could change the reserve once more
facts are known
Constraint: Reserves should be based on the facts provided
plus standard ranges for similar losses. Do not invent
specific medical bills, repair estimates, or jury verdicts.
Round to a reasonable number.
Loss: [paste FNOL + any known facts]
Chaining the Triage Workflow
The real power comes from chaining these prompts. A simple workflow:
- Run the coverage tier prompt
- Run the severity scoring prompt
- If severity is 3 or higher, run the routing prompt with the major-loss team list
- Run the initial reserve prompt
- Generate a triage memo combining all four
A combined triage memo prompt:
You are a claims supervisor producing a triage memo from
the FNOL below. Output a structured memo with these
sections:
- Coverage tier (Covered / Not Covered / Ambiguous + 2
sentences)
- Severity (1-5 + confidence + 3 drivers)
- Recommended team and routing
- Initial reserve recommendation with breakdown
- Special handling needed (SIU, attorney, executive review,
none)
- 5 follow-up questions for the assigned adjuster
- Open items to investigate
Constraints: Use only facts in the FNOL. Do not invent
medical specials, repair estimates, or policy language.
FNOL: [paste]
This memo is what your assigned adjuster sees first thing in the morning. It collapses what used to be 20 minutes of triage into 60 seconds of AI generation plus a few minutes of human review.
Beyond Single Claims: Batch Triage
For carriers handling hundreds of FNOLs per day, AI can batch-process. With ChatGPT Team, Claude Team, or your carrier's procured AI platform, you can build a workflow that:
- Pulls overnight FNOLs from the claims management system
- Runs the triage memo prompt on each
- Outputs a daily triage worksheet for the supervisor
- Flags any FNOL where AI confidence is low for human triage
The integration piece is for IT, but the prompts you have built can be the foundation.
Catastrophe Triage
When a hurricane, hailstorm, or wildfire generates thousands of FNOLs in 48 hours, triage becomes the bottleneck. AI can help in three specific ways:
- CAT classification. Identify which claims belong to a CAT event vs. unrelated daily claims.
- Severity prediction at scale. Apply the severity prompt to every CAT FNOL.
- Field adjuster routing. Group claims by ZIP code for efficient field deployment.
A useful CAT prompt:
You are a CAT claims coordinator for [STORM NAME] in [STATE].
Below are 50 FNOL summaries. For each:
- CAT YES/NO
- Severity 1-5
- Inspection priority (Urgent / Standard / Low)
- Recommended team
Output as a CSV with columns: claim_number, cat_yes_no,
severity, priority, team. No commentary.
FNOLs: [paste]
A Reality Check
AI triage is a powerful productivity tool, not a substitute for licensed supervisor judgment. Several state DOIs are paying close attention to how AI is used in claims handling, and the bad-faith bar is watching too. Always:
- Have a licensed supervisor review and approve any AI-recommended routing or reserves
- Keep an audit trail of what the AI recommended vs. what the supervisor decided
- Escalate any low-confidence or ambiguous output to human-only triage
- Keep your triage prompts current with carrier appetite, team structure, and regulatory changes
Key Takeaways
- The triage decision stack — coverage tier, severity, routing, reserves, special handling — can each be augmented by AI prompts.
- Chaining the four prompts produces a comprehensive triage memo in seconds, which a supervisor reviews in minutes.
- Batch and CAT triage are where AI productivity gains are most dramatic, but require IT integration.
- All AI triage output is a recommendation, not a decision. Licensed supervisors confirm and document.