Return to page

Progressive Uses H2O Predictive Analytics for UBI

 

 

 Progressive is one of the largest auto insurers in the United States with over 13 million policies in force. Progressive is a pioneer in data analytics with more than 14 billion miles of driving data collected through its telematics offering, Snapshot. Hear from Pawan Divakarla, Data and Analytics Business Leader, on how H2O is helping build models and derive insights in just seconds using open source Machine Learning.

 

Talking Points:

Speakers:

Pawan Divakarla, Data and Analytics Business Leader, Progressive

Edward Agarwala, Director Of Analytics, Progressive

Read the Full Transcript

 

 

 

Pawan Divakarla:

 

Progressive is a data driven company and we deploy our learnings from data using technology as a way to ensure our end users.

 

Data Sources

 

Edward Agarwala:

 

We have over 14 billion miles that have been collected in our customers driving data. That's further than Voyager has traveled out of the solar system. We also have all sorts of new sources of data that are coming in. We are testing, collecting GPS data from some of our customers and we're also testing mobile as a device. This is collecting speed according to the GPS as well as accelerometer data from your phone to be able to replace the device.

 

Utilizing H2O

 

Pawan Divakarla:

 

We were collecting a lot more data. It was coming to us at a much faster pace. One area where we were seeing a pain point was our time to insight and we decided to use machine learning algorithms as a way to better understand the data so we could make predictions about what's happening in the insurance marketplace. Having H2O as part of our data science platform gives us a much shorter time to insight. So now folks are able to do many models in much shorter periods of time, and what historically was a bottleneck where we couldn't entertain other lines of business and their data science needs, we can actually address their data science and predictive modeling needs now because we have a much faster throughput of our models and the business value we're able to generate from it. Before big data, we could only possibly look at a state at a time or a few states at a time or a region at a time, but now we can actually look at all states and we can look at countrywide predictions and we can use more data at scale that in the past we couldn't. And H2O is obviously piece of this puzzle now that we can actually use it more at scale.

 

Memory Performance

 

Edward Agarwala:

 

H2O has great in memory performance, even on single nodes. It will utilize all of the cores even on single nodes. It has all of the latest and greatest algorithms including stochastic GBM, including Random Forest, including GLM. And it has things like the weights, it has the different loss functions. There isn't really a single package or instrument that has all of these things readily available.

 

Future Plans with H2O

 

Pawan Divakarla:

 

We have many plans to use H2O and machine learning and predictive analytics across the different business units we have here for customer churn, for retention analysis purposes, for billing, for fraud detection, even for threat analysis and so on, where we see a lot of data coming to us and we need to analyze it in shorter periods of time and actually make sense of it. In the past, we would have to have really sophisticated models on small data, but now we can actually have lots of sophisticated models on big data, which gives us a whole new dimension of new insights that we didn't have in the past. We're a consumer brand and our brand is growing and our mobile traffic is growing, our web properties are growing. Snapshot, our usage-based product is also growing. So all of these contribute more data and my goal is to derive more insights out of this data.

 

Predictive Analytics

 

Predictive analytics is making a very positive culture shift for us in in the company, and I see it growing almost exponentially as we go into the next few years. Now, as we hire new data scientists, they actually have a platform that's very mature and H2O is part of the platform. They can take their curiosity and actually run these algorithms at scale and then present new scenarios that we may not have thought in the past. I think it's fostering a lot more creativity amongst our analytic population and our data science population. Our data scientists and advanced analytic users are like craftsmen, and now we're giving them the right tools that they can actually practice their crafts. Giving that empowerment to our data scientists is key for us. We are a insurance company and we are constantly assessing risks and one way of assessing risk is the predictiveness of it. So predictive analytics has always been a core component of what we do, but I see it growing quite rapidly over the next few years. We'll have more data, we'll have more signals, and we need to better understand the data. H2O is like an enabler in how people are thinking about the data and how they want to use the data, and that's come in very handy for some of our data scientists and advanced analytic users. Now they have the right tool set that they can use on the data.