Zurich Uses External Lidar Data to Deliver More Transparent Policy Pricing and Meaningful Underwriting Insights

If you’ve ever had to insure anything, you’ve likely come across Zurich Insurance Group. Offering insurance products for a range of businesses and public sector entities like governments, schools and hospitals, Zurich has grown to become one of the largest insurers.

Like all insurers, Zurich is in the business of using data to understand risk and underwrite insurance products. But while there is plenty of data available on individuals and assets, it often comes in different forms — meaning its impact can be limited when assessing risk and setting prices.

As a longstanding Snowflake customer, Zurich used the AI Data Cloud to turn a UK government data set covering over 300 billion data points into a usable form that automates elements of the underwriting process, creates more transparent pricing and attracts new customers.

Data has been the lifeblood of insurance for decades now. But the data sets available, and the way they can be used, have shifted in recent years. “There’s now a move to make the most of the data rather than just keeping it,” explains Jonathan Davis, Data Science Lead at Zurich. “There’s also a growing focus on external data. There’s a lot of it out there that’s freely available, but it’s often difficult to use.”

Of particular interest to Zurich’s UK operations was the government’s national lidar (light detection and ranging) data, which shows the height of buildings and their surroundings in England. The potential use cases for this data are vast, allowing Zurich teams to see how tall buildings are, the height and distance of nearby trees and other physical risks that could influence underwriting decisions.

But with billions of data points, Zurich’s previous infrastructure lacked the throughput to make use of this data in a reasonable timeframe. “It could be usable,” says Davis. “We tried loading it the old fashioned way in Python and on a general SQL server but it failed very quickly. Even with a proof of concept project on just one of the 6,500 files, there was no way we could scale it.”

Zurich had already been using Snowflake for a number of years to ingest and process underwriting data. And with the AI Data Cloud, Davis and his team found they had the scale, throughput and a range of other features to bring the use case together.

“The native geospatial support was essential,” explains Davis. “We’ve also benefited significantly from support for Python in Snowpark. Our team is familiar with it, and it’s great to work in local development environments and push it to Snowflake — and everything just works seamlessly.”

Beyond the familiarity of Python, native support for different languages offered the team access to crucial pre-made libraries that made the use case possible. “LAZ and LAS files are commonly used in lidar data, and there’s not much support out there for reading these filetypes,” adds Isaac Brocklesby, Data Scientist at Zurich. “But there are two really well-made packages in Python for reading these. If we didn’t have access to them, the whole use case would’ve been a lot more tricky.”

While the use case has been a long time coming for Zurich, the actual processing was fast once the team set the AI Data Cloud on the task. In just 100 hours running a medium-sized Snowpark warehouse, everything was processed and ready. Since preparing the data set, the benefits have been significant across Zurich’s consumer and enterprise underwriting.

“All types of underwriting can benefit from a greater understanding of risk,” says Anna Collins, AI Data Lead at Zurich. “Now we can see exactly what buildings are high or mid-rise and can better assess the risk of flooding or wind damage.”

And with exact height data of every building and every area around a building, the insurer can also calculate the risk of tall trees falling on a property or how much of a property has a flat roof that could degrade over time. “We’ve been able to develop some complex algorithms to turn all this data into proper risk analysis,” explains Brocklesby. “Relating two points spatially can be computationally expensive, but Snowflake’s H3 feature allows us to use spatial indexing to calculate the relations between height. It’s much faster and was a cornerstone to unlocking value from this data set.”

With more insights at its disposal, Zurich can protect itself against insuring risks that could lead to significant losses. But it can also more accurately price individual cases, potentially attracting new customers who are being miscategorized and offered more expensive policies by other providers. 

And when customers are on board with Zurich, they get a better experience with an insurance provider that better understands their situation. “It’s a faster service now,” says Davis. “We don’t have to ask as many questions during the quote and onboarding process. I’m sure some of our customers used to think ‘you should already know this!’ and now we do.”

Accurate lidar data also leads to more efficient underwriting processes at Zurich. “It’s part of our automation journey,” explains Will Davis, Lead Machine Learning Engineer at Zurich. “In the past, underwriters would often have to use Google Street View to try and estimate the heights of different buildings. This could take minutes at a time for each address. Now it’s near instant. And for a portfolio of thousands of properties, that saves a lot of time.”

Zurich has already found significant benefits from ingesting and transforming the national lidar data set. But it’s just getting started. “It’s an absolute mine of data,” says Davis. “Spend enough time there and you’ll find even more gold nuggets. There’s a lot of enrichment we still want to do.”

The team also has eyes on offering the lidar data as a self-serve platform for anyone within Zurich. Davis is looking into Snowflake’s Streamlit support as a way to develop a native app around the lidar data set. And the team is considering how to deploy AI and machine learning models in Snowflake so it can create a managed layer for users to run different models as well.

It’s not just platform features that have the team excited for the future, though. Davis and the rest of the data science team at Zurich also look forward to continued support from Snowflake that helps make new use cases possible. “Snowflake’s team even added support for some of the Python libraries we needed,” says Davis. “The platform is powerful, no doubt. But it was the support we got that has really made the difference.”