10 Million Data Points 
365 Days Before Katrina
See Around Corners.
We help (re)insurance industry participants more accurately calculate Expected Loss through Big Data.
Datum Cos. is a predictive analytics technology company serving the insurance and reinsurance industry.  The Datum team includes data scientists, physicists, mathematicians and insurance industry experts all working together to develop applications to solve important industry problems using Big Data technologies.
Understanding Property Catastrophe Event Risk Through Big Data: Eurus® 
Our machine learning models analyze billions of historical weather feature data points representing Terabytes of data on our high performance computing clusters to understand complex interrelationships of weather parameter readings.  Our models calculate catastrophic event probabilities up to one year in the future which enhance the results of legacy stochastic dynamical models.  Learn more about our technology.
Why is a new view of Expected Loss important?
The industry is currently flying blind.
Catastrophe models in wide use throughout industry today utilize stochastic synthetic event catalogs to create event distributions which inform vulnerability models to estimate Expected Loss.  While these models provide some measure of risk assessment, they are not sufficiently accurate and therefore do not provide a complete view of risk.  Some of the drawbacks of legacy models include:

  • Poor track record estimating medium term industry catastrophic loss
  • Consistently large Expected Loss standard deviation
  • New catastrophic events render previous catalogs less reliable 
  • Frequent model revisions "rewrite" the software and reset PMLs
  • Limited "modeled loss" perils and geographic regions
  • Industry applies multiples to Expected Loss effectively disregarding model output
We believe there is a better way.  

In our view, legacy models do very well assessing vulnerability and monetary loss should events occur, but struggle when assessing hazard risk (event probability).  Legacy models estimate hazard risk based on catalogs of thousands of simulated events.  

We analyze billions of actual weather parameter readings to construct our predictive models. When combining legacy model results with Big Data predictive models, the result is a more robust, multi-model, more accurate view of medium term Estimated Loss  Learn more about our tools.
Enhance Expected Loss Accuracy
We apply our machine learning models to recalibrate legacy property catastrophe model exceedance probability curves using our Big Data platform to create a secondary EP curve providing a new view of PML and a more accurate pricing assessment.
Improve Unmodeled Loss Visbility
Legacy property catastrophe models are built around synthetic event catalogs in specific regions covering specific perils. As we model actual weather feature data continuously and globally, we are better geared to deliver probabilities of a variety of catastrophic weather events anywhere in the world.
Create New Product Opportunities
We have the capability to lever our state of the art cloud based computational platform and Big Data machine learning engine to analyze billions of data points across geographies, perils, and lines of business to price new types of risk as new product opportunities come to market.