Demyst is an external data access platform that enables clients to push data into use-cases. The below is a conversation between John Siegman, Co-Founder of HazardHub and Prashant Reddy, Demyst's Head of Data Advisory.

Prashant: John, tell me a bit about HazardHub’s core value prop.

John: At HazardHub we believe that our customers should know as much as possible about a location and that price should not be a hindrance to that knowledge.  That's why we’ve productized more than 900 variables, with data ranging from perils, to property characteristics, to permits, and we make it all available at one very affordable standard price point.

Prashant: That’s quite a bit of data, do we really need 900 attributes?

John: The better question is, why not make even more data? There are constant improvements to both data and models, yet the industry runs on relatively little data. While it’s been a standard practice to make the insured or the agent answer a bunch of questions, those days are rapidly coming to a close.

In truth, we don’t always know the best data that is indicative of claims. Interestingly most insurers also do not seem to know the best data, either. They know what causes claims from the data that they’ve been exposed to, but many times they’re relying on data elements that were created 10, 20, or sometimes 30 years ago. For example, did you know that the most popular fire protection tool was developed based on a study from 1971?  (Hint – 5 miles from the fire station isn’t valid anymore.)

Prashant: Great – HazardHub provides data for claims (that’s pretty well known at this point) – what other use-cases is the data useful for?

John: HazardHub provides data to avoid claims. Companies need a ton of data to do prefill, marketing, actuarial analysis, claims validation, in addition to underwriting and rating.  These types of information tend to be niche and expensive.  We try to make all of that information easy to access and cost effective to use.

Prashant: There are some pretty strong incumbents in the world of property data. How would you recommend an insurer get comfortable with your scores vs. something from CoreLogic or Verisk?

John: How can an insurer get comfortable with the data? Test it. But not in the way you might be thinking. A lot of people think that data sets are comparable.  If an insurer already has a wildfire model, for example, and they look at ours, okay, checkbox for each provider.  No net gain to the insurer.  That’s a bad test. 

We see the best results when insurers - adopt a challenger/champion testing methodology. The current whatever is the champion. The outside whatever is the challenger.  Test one against the other.  Let the best data model win.  We see a lot of testing where the insurer wants to see if a single variable will provide additive lift to their existing model.  That’s a recipe to justify what currently works, because it is quite rare that a single variable can make a significant change.

Prashant: Can you give an example of a test that didn’t work?

John: For example, we recently participated in a test where the insurer tried to improve upon their existing fire rating methodology by looking at an additive variable from our data set.  It didn’t work – and should not work - because it’s a poorly designed test. What they missed is that we identified over 40,000 “F” wildfire risks – which contain 88% of actual wildfire losses - that they did not know that they were insuring and over 30,000 “F” hurricane risks (which represent 95% of hurricane claims) that they were insuring. 

Prashant: As a member of the Demyst team, I have to point out here that we’ve assisted our insurance clients with dozens of unconstrained loss modeling tests, and in the world of risk, what we see is that we need to focus not only on lift, but on improve accuracy in classification (or really misclassification in John’s example).

Any additional thoughts John?

John: The world of data is changing. Datasets are better, more powerful, and less expensive than ever before. You can build incredible customer experiences using highly reliable, economic 3rd party data from companies like HazardHub. All it takes is just a bit of tweaking to your current modeling efforts.

Prashant: And that’s a wrap – actuaries, data scientists, data analysts, and external data professionals – come to Demyst to access external data from HazardHub, and a hundreds of other consumer and commercial data sources to optimize your workflows. Thanks for your insights John!