Dan Marzouk: How Aegis Insurance is Overcoming Data Discrepancies to Price Catastrophic Risk

Dan Marzouk, Senior Vice President of Data Science at Aegis,
explains how data science is shaping their approach to insurance

Wildfires are difficult to predict, grow rapidly and have the
potential to cause damage worth tens of billions of US dollars.


This is a problem for insurers trying to price risk. The
solution? Using data to develop a more complete understanding of
risk, argues Aegis Insurance Senior Vice President of Data
Science Dan Marzouk in this week’s episode of the Business of
Data Podcast.


When evaluating, for example, the chances of a wildfire affecting
a suburban home, there are a wide range of data points to
consider and a variety of data sources to include. However, not
all sources are of equal quality.


“The challenges are similar to comparing a Google review, a Yelp
review and a Facebook review for a business. Each of those
[reviews] have their pros and cons,” Marzouk notes. “Each of our
data sources also have their pros and cons.”


The differing quality of data sources can lead to discrepancies
in the data. That’s where data science comes in. Creating a
consistent risk assessment requires building a model that
quantifies the accuracy of input data.


“Over time we start to learn and utilize what we think is
accurate from one dataset and continue on that path to build our
own data integration system that understands what we believe to
be the most accurate system,” says Marzouk.


Of course, weighing tens of thousands of data points takes time.
However, as Marzouk explains, in the age of instant everything it
is crucial to provide insights to decision-makers quickly.


“To do that, we have to both understand how to aggregate that
data quickly and cull out what’s not as important or useful,” he
says. “And be able to develop something that the underwriter can
make a decision on quickly.”


Ultimately, to meet the business need the data must help to
create a product that is appealing to the customer. That means
that data scientists must also maintain a commercial awareness.


“Customers don’t buy things because you told them that the model
says [they’re] going to buy it,” Marzouk quips. “That’s my advice
to the data science community. Take a step back and say, ‘I know
the data’s telling me this, but does it make sense?’”
Key Takeaways


Understand the data you have. Is it granular
enough? How reliable is the source? The answers should inform
your model.


Maximize your data points. Innovative
technologies like image recognition can dramatically increase
the number of data points available


Take a step back. Remember to evaluate what
the data is telling you in the light of all other available
information