LogoInsurAItools
  • Reviews
  • Free Tools
  • Solutions
  • Categories
  • Compare
  • Glossary
  • Blog
  • Pricing
LogoInsurAItools
← Back to Glossary

Claims Severity Model

A model predicting the ultimate cost of an individual claim, used to set reserves, prioritize handling, and flag high-exposure files.

technicalPublished 2026/06/07Last verified 2026/06/07

FAQs

At what point in the claim lifecycle is a severity model most valuable?
Severity models have the greatest operational impact when applied early — at first notice of loss or within the first 30 days — before the claim's development trajectory is established. Early identification of high-severity claims enables proactive interventions that are far less effective once litigation is filed or medical treatment has advanced.
Can a severity model replace the adjuster's reserve judgment?
No, and most actuarial and regulatory standards do not support fully automated reserving on individual claims. The model provides an anchor and a flag for outliers, but the adjuster retains authority to set the reserve based on their assessment of specific claim facts. The model's value is in focusing adjuster attention and establishing a defensible initial reserve.
How do we validate a severity model for regulatory purposes?
Validation should demonstrate predictive accuracy on an out-of-time holdout sample, show that model predictions are well-calibrated (not systematically biased high or low by jurisdiction or claim type), and confirm that the model does not produce discriminatory outputs on protected class proxies. Results should be documented in the model's governance record.

Related Terms

  • Model Drift

    Degradation of a deployed model's predictive accuracy over time as input feature distributions or outcome relationships shift from the training environment.

  • SIU Referral

    The process of routing a suspicious claim to the Special Investigations Unit for investigation of potential fraud before settlement.

  • Claims Leakage

    Measurable overpayment on claims relative to the theoretically correct settlement, resulting from process failures, errors, or inadequate investigation.

  • IBNR Reserve

    Incurred But Not Reported reserve: a liability estimate for losses that have occurred but have not yet been reported to the insurer.

Related Items

  • Charlee.ai

    Predictive analytics for claims litigation

  • Gradient AI

    ML for underwriting risk and claims optimization

  • Five Sigma

    AI claims management with adjuster decision support

LogoInsurAItools

Independent AI tool reviews for insurance agents and brokers

Product
  • Reviews
  • Free Tools
  • Solutions
  • Categories
  • Compare
Resources
  • Glossary
  • Blog
  • Pricing
  • Search
  • Collection
  • Tag
Company
  • About Us
  • Privacy Policy
  • Terms of Service
  • Sitemap
Copyright © 2026 All Rights Reserved.

A claims severity model is a predictive model that estimates the ultimate total cost — indemnity, medical, and legal expense — of an individual claim, typically at or shortly after first notice of loss, using available claim characteristics, claimant data, policy information, and external enrichment data.

How it works / Why it matters

Severity models address a fundamental operational challenge in claims management: the final cost of a claim is unknown at opening, yet decisions about reserve levels, adjuster assignment, litigation strategy, and settlement authority must be made throughout the claim's life. A model that can accurately rank claims by expected ultimate cost allows claims organizations to concentrate resources on high-exposure files before they develop adversely.

Severity model architectures in insurance include:

  • Regression models (generalized linear models with Tweedie or gamma distributions) that estimate expected cost as a function of structured claim attributes: line of business, cause of loss, claimant age, injury type, jurisdiction, and represented/unrepresented status.
  • Gradient boosting models such as XGBoost or LightGBM that capture complex non-linear interactions between claim features and ultimate cost, particularly in bodily injury lines where severity drivers interact in ways linear models cannot capture.
  • Neural network architectures applied to high-dimensional inputs including nlp-submissions-extracted text from adjuster notes, medical reports, and attorney correspondence.
  • Ensemble models combining multiple base models to improve stability and reduce prediction variance.

The model output — a predicted ultimate cost or a severity tier (low, medium, high, catastrophic) — is used by the claims system to set an initial reserve, route the claim to the appropriate adjuster authority level, flag files for medical management or litigation management, and trigger SIU referral criteria from siu-referral rules.

In practice

A workers compensation carrier might deploy a severity model that scores every new indemnity claim within 24 hours of opening. Claims scoring in the top decile are automatically assigned to senior adjusters, flagged for early attorney intervention, and reserved at the model's predicted ultimate value subject to adjuster review. This approach reduces claims-leakage by ensuring that high-exposure claims receive proportionate attention from day one.

Charlee AI, Gradient AI, and Five Sigma offer pre-built severity scoring models that integrate with major claims management platforms. RapidClaims applies severity prediction specifically to medical bill review.

Model output must be monitored for model-drift as the claims environment — medical costs, legal climate, social inflation — evolves. Reserve adequacy analysis routinely compares model-predicted ultimates against actual paid development.

Related concepts

See model-drift for how severity models degrade as the claims environment changes, and case-reserving for the manual reserving process that severity models augment.