Complete AI Training
Sign inGet my AI kit

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

AI agent for insurance risk analysts

Claims Severity Spike Investigator Agent

Explain a claims severity spike with tested evidence within a day or two.

AnalysisFinanceInsurance
See how it worksGet access
Claims Severity Spike Investigator Agent: what goes in, what the agent does and what you get

What it does

Average claim cost jumps one quarter and management asks why, but the answer takes weeks of digging. This agent monitors average severity and flags a spike when it passes a set level. It splits the jump by cause of loss, region, vendor, product and accident period, and tests each possible explanation against the data, such as a few large claims, a new repair vendor, inflation in parts or a change in coverage mix. It rechecks the strongest explanation on a holdout period to see if it holds. If it does not hold, it moves to the next. It drafts findings and what to do next. The analyst approves before they go to management.

How it works

Follow the arrows from top to bottom. The orange dashed arrow is the loop: when a check fails, the agent goes back and tries again.

Start and resultWhat it doesA check on its own workWaits for your OKGoes back and retries
Yes, continueApprovedNo 1 STARTS WHEN Severity alert fires 2 USES A TOOL Pull claims and the severity history 3 DOES Confirm the spike is not caused by a few very largeclaims 4 DOES Split the increase by cause, region, vendor, productand period 5 DOES Test each possible explanation against the data 6 DOES Rank explanations by share of increase explained 7 USES A TOOL Test the top explanation on a holdout period 8 CHECKS THE RESULT Does the explanation hold on the holdout period? If not: drop it and test the next ranked explanation.Back to step 6. 9 DOES Draft findings with evidence and suggested actions 10 YOU APPROVE Analyst approves before the findings go tomanagement 11 RESULT Severity investigation note
Read the steps as a list
  1. Severity alert fires
  2. Pull claims and the severity history
  3. Confirm the spike is not caused by a few very large claims
  4. Split the increase by cause, region, vendor, product and period
  5. Test each possible explanation against the data
  6. Rank explanations by share of increase explained
  7. Test the top explanation on a holdout period
  8. Does the explanation hold on the holdout period?If not: drop it and test the next ranked explanation. Back to step 6.
  9. Draft findings with evidence and suggested actions
  10. Analyst approves before the findings go to managementThe agent waits here for your OK.
  11. Severity investigation note

How it decides

It ranks explanations by how much of the increase they account for, and keeps one only if it also holds on data it did not use to find it.

  • Alert when severity is 10 percent above the 12 month average
  • Treat claims above $250,000 separately
  • Accept an explanation covering at least 50 percent of the increase
  • Test every explanation on a holdout period

Make it yours

Every agent is a starting point. You choose these settings for your own situation.

  • Alert threshold (default 10 percent)
  • Large claim cutoff
  • Dimensions to split by
  • Holdout period length

What keeps you in control

It always asks you first

  • Analyst approves findings before they go to management

Hard limits

  • Never changes reserves or claims
  • Reports an unexplained part honestly

It stops when

  • Done: explanation confirmed or stated as not found
  • Stop: claims data for the period is incomplete

Set it up

We guide you through the set-up, step by step

Members get the full set-up guide for this agent. No technical skills needed: you copy, paste and upload.

10 minto set it up in your AI
5 AIsChatGPT, Claude, Copilot, Gemini, Grok
  • One set of instructions to paste into your AI, with the clicks for ChatGPT, Claude, Microsoft 365 Copilot, Gemini and Grok
  • The agent then walks you through connecting your own data, one source at a time
  • A downloadable copy with the flow chart, the rules and the full guide
Get access to this agent

An example run

What happensOctober severity was $6,420 against a 12 month average of $5,600. The agent found 4 large claims explained 9 percent of it. Splitting by vendor, a new repair shop in one region accounted for 48 percent, but on the holdout month it explained only 15 percent, so it failed. Parts inflation explained 61 percent and held on the holdout. The analyst approved the note.

More agents for insurance risk analysts

Data cleanup · Quality control

Insurance Exposure Unit Reconciliation Agent

Exposure schedules in consistent, defined units.

8 steps2 checks2 jobs
Data cleanup · Analysis

Loss Run Analysis Agent

A clean, checked loss history with frequency, severity and large losses

10 steps2 checks3 jobs
Analysis · Risk management

Mule Account and Network Link Analysis Agent

Find the accounts linked to a confirmed fraud account and rank them by strength of evidence.

10 steps1 check3 jobs
Risk management · Analysis

Climate Scenario Loss Modeling Agent

Show how climate scenarios hit the actual portfolio and which mitigations work.

11 steps2 checks3 jobs