Dual GLM Actuarial Engine
Motor insurance priced by actuarial mathematics, not generic averages
We decouple accident likelihood (Poisson Frequency) from claim cost (Gamma Severity), producing personalized, transparent, and auditable motor premiums.
Live Sample Rating
Model GLM-FR-V1Calculated Final Premium
€340/ year
783,573
policies modelled
46,090
claims observed
Dual GLM
Poisson × Gamma pipeline
22
French regions rated
Interactive Simulation
Test the pricing sensitivity in real time
See how small shifts in driver age, vehicle power, and no-claims bonus change the expected annual loss and premium.
Live Actuarial Sandbox
Instant Risk & Premium Estimator
Tweak the levers below to inspect how driver age and claims history shift the loss equation.
Pure Expected Loss: €75
The Modeling Pipeline
Three rigorous stages behind every underwriting decision
01
Predict claim frequency (Poisson GLM)
A Poisson regression with log-exposure offset estimates the annual likelihood of a policy filing a claim, incorporating driver age, bonus-malus coefficient, and territorial density.
02
Predict claim severity (Gamma GLM)
A Gamma regression model with log link predicts the expected monetary cost per claim condition upon an accident occurring, sensitive to vehicle horsepower class and repair brand groups.
03
Pure premium & technical loadings
Pure Expected Loss = Frequency × Severity. The engine loads commercial expenses, taxes, and margin buffers, enforcing a stable actuarial minimum floor.
Actuarial Precision
Two drivers, two tailored premiums
Traditional insurers charge the group average. Our model isolates idiosyncratic behavioral and demographic factors.
Policyholder A
High Risk21 · Novice Driver · Hot Hatchback
Policyholder B
Low Risk45 · Experienced Commuter · Family Sedan
Policyholder B saves over 92% on the same underwriting baseline because actuarial models price risk exposure, not arbitrary market markups.
Platform Capabilities
Built for retail quotes and institutional portfolio re-rating
Deployable for direct-to-consumer digital brokerages, underwriter workbenches, and reinsurance capital modeling.
High-Throughput Batch Rating
Price complete insurer portfolios containing 100,000+ policies in seconds using optimized vectorized Pandas pipelines.
Continuous Retraining Flywheel
Anonymized quotes automatically feed Supabase persistence tables, providing clean research data for longitudinal drift analysis.
Geographic Stress Simulation
Stress-test policy books against urban congestion shocks (e.g. traffic surges and severe weather claim anomalies).
Solvency II Audit Trail
Zero black-box risk. Pure premiums, relativity multipliers, and loadings are fully decomposable for regulatory reporting.
Frequently Asked Questions
Actuarial methodology & compliance
Clear answers to how the statistical models, data telemetry, and rating algorithms function.
In modern actuarial science and under regulatory regimes such as Solvency II and European insurance laws, pricing models must be transparent, auditable, and monotonic. Standard deep neural networks can produce non-monotonic pricing artifacts (e.g. charging an older driver randomly less than a slightly younger one without statistical justification). By separating claim frequency (Poisson regression with log-exposure offset) from claim severity (Gamma regression with log link), actuaries maintain full mathematical interpretability while achieving high statistical predictive power.
Ready to Price
Calculate your personalized rate today
Enter policyholder variables for an instant quote, or upload a complete insurer CSV portfolio for batch pricing.