Insurance Pricing Model Modernisation
EdgeRed built 50+ production-grade pricing models on Databricks for a major Australian insurer — replacing legacy GLMs with Gradient Boosting Machine techniques across a complex, multi-product portfolio, achieving a 4-6% uplift in predictive accuracy that translates to materially more reliable cost forecasts and stronger pricing decisions at scale.
The Challenge
A major Australian insurance provider needed to modernise the pricing models underpinning a large, diverse portfolio covering multiple product types and risk categories. Existing Generalised Linear Models were reaching the limits of their predictive capability — unable to capture the complexity of risk patterns across hundreds of variables and a diverse product range. The business needed more accurate, data-driven pricing to remain competitive and price risk more sustainably. Legacy models fragmented across platforms were also inconsistent, difficult to scale and limited the team’s ability to test assumptions or model scenarios efficiently.
Technology Used
- Databricks — ML platform; GBM model development, feature engineering pipelines, and reproducible model training at scale
- Streamlit — interactive scenario modelling application for actuarial and pricing teams to adjust assumptions in real time
- Power BI — model performance tracking and feature insight dashboards for ongoing production visibility
What We Built
EdgeRed built 50+ advanced predictive models using Gradient Boosting Machine techniques on Databricks — replacing the legacy modelling framework with a modern, scalable approach across the full portfolio.
-
Advanced model development
GBM models built across 56 products — analysing hundreds of variables across geography, policyholder and product dimensions to identify key drivers of claim frequency and severity with far greater precision than legacy GLMs -
Feature engineering & pipelines
Robust data pipelines and feature engineering developed to support reliable, reproducible model training at scale — ensuring consistent model quality across the full portfolio -
Interactive scenario modelling
Interactive Streamlit application built allowing actuarial and pricing teams to dynamically adjust key assumptions and model scenarios in real time — without requiring engineering support -
Performance tracking & governance
Model outcomes and key feature insights surfaced in Power BI — giving actuarial and pricing teams ongoing visibility over model performance and pricing decisions in production
Augmented Services
EdgeRed supplied a dedicated onshore team to extend the client’s internal data capability – without the overhead of a full-time hire. The team brought a blend of seniority and experience – 1 ML Engineer and 2 Data Scientists.
What that means in practice:
- Senior capability on demand - right skills, right time, no long-term headcount commitment
- 100% onshore delivery - every person on your project is based in Australia
- Flexible resourcing - scale the team up or down as project needs change
- Deep bench - draw on the collective knowledge of a 50+ person consultancy, not just the people in the room
The Outcome
50+ production-grade pricing models delivered across the full portfolio — achieving a 4-6% uplift in predictive accuracy that translates to materially more reliable cost forecasts and stronger pricing decisions at scale. The transition from legacy GLMs to GBMs enabled more tailored risk assessment by capturing complex patterns that previous models couldn’t detect — giving the business a defensible, data-driven basis for pricing across a portfolio of significant commercial value. Actuarial and pricing teams can now test assumptions and model scenarios in real time via a Streamlit application, without requiring engineering support.
