Generalized Linear Models for insurance pricing: Poisson, Gamma, and Negative Binomial
GLM (Generalized Linear Model) is a statistical technique used in insurance to predict claim costs and set appropriate premiums. It helps answer questions like: "How much should we charge a 30-year-old male driver?" or "What's the expected claim amount for a homeowner in California?"
Instead of charging everyone the same price, GLM allows you to adjust premiums based on risk factors like age, location, coverage type, and more—making pricing fairer and more accurate.
Best for: Predicting the number of claims (frequency)
Example use case: "How many auto insurance claims will we receive from drivers aged 25-30?"
Why it works: Poisson distribution is designed for count data (0, 1, 2, 3 claims...). It assumes that claims are rare events that occur independently.
Best for: Predicting the average cost of claims (severity)
Example use case: "What's the expected dollar amount when a claim occurs?"
Why it works: Gamma distribution handles positive continuous values (claim amounts) and naturally models right-skewed data (most claims are small, few are large).
Best for: Predicting claim counts when there's high variability (overdispersion)
Example use case: "How many claims for customers with very different risk profiles?"
Why it works: Like Poisson but allows for more variance. Better when some policyholders have way more claims than others.
Gather past claims data including: number of claims, claim amounts, policyholder characteristics (age, location, coverage type), and exposure (time insured).
Choose variables that might affect claims: age, gender, location, vehicle type, credit score, etc. ADaaS helps identify the most important factors automatically.
ADaaS fits Poisson (for frequency), Gamma (for severity), or Negative Binomial models to your data. The model learns how each risk factor affects claims.
Use the model to predict expected claims for new policies. Multiply frequency × severity to get expected total cost, then add profit margin and expenses to set premium.
See exactly how much each risk factor affects the outcome. For example: "Being male increases expected claims by 15%"
Visual ranking of which variables matter most for predictions. Focus on the factors that truly drive claims.
AIC, BIC, deviance, and other statistics to evaluate how well your model fits the data.
Visualize how well your model predictions match reality. Identify areas where the model might need improvement.
Download comprehensive PDF/DOCX reports with all model results, charts, and interpretations for stakeholders.
Goal: Set fair premiums for auto insurance based on driver characteristics
Step 1 - Frequency Model (Poisson): Predict how many claims each driver will file
Result: Young drivers (age 18-25) have 2.3x more claims than older drivers
Step 2 - Severity Model (Gamma): Predict average cost when a claim occurs
Result: Urban drivers have $1,200 higher average claim costs
Step 3 - Calculate Premium: Frequency × Severity × Profit Margin
Result: 22-year-old urban driver: $1,800/year vs 45-year-old rural driver: $650/year
Set risk-based premiums for auto, home, health, or commercial insurance products.
Identify high-risk vs low-risk customer segments for targeted underwriting.
Analyze whether current rates are adequate or need adjustment based on recent claims.
Generate actuarially sound rate filings with statistical justification for regulators.
Upload your claims data and get instant pricing insights