(Image credit: Shutterstock.)
The concept of “predictive analytics” has been around in the insurance industry for a long, long time—way before “big data” was a thing. In the early days, when intelligent data hid in filing cabinets and spreadsheets, predictive analytics was essentially an insurance professional slicing and dicing data to make decisions based on which policies were costing the most money. Now, of course, predictive analytics is much more sophisticated, leveraging algorithms to try and more consistently pinpoint accurate outcomes.
But for most insurers, there’s still room for improvement. The influx of data—both structured and unstructured—is a goldmine that is mostly still untapped. And, without diving into it and making it consistently useful and actionable, insurers can’t become data- and analytics-driven organizations, which is the only way to true business success in the 21st century.
So, how can today’s insurers take advantage of predictive analytics, and harness all the data that the connected world has to offer? What must they do to adapt and succeed in times of rapid change?
Made for Each Other: New Data Sets and Machine Learning
For a data-rich industry like insurance, the ability to predict outcomes or customer behavior is invaluable because it directly affects insurers’ cost of doing business. Imagine the benefits of knowing the answers to questions like, “What is the likelihood that X customer will have losses?” and “How big will those losses be?”
Answers to these questions among others are fundamental to the success of an insurer, and they can be found in data. But effectively collecting and mining data for predictive intelligence depends largely on the predictive model used. A common predictive analytics model used by many insurers even now is the generalized linear model (GLM). Unfortunately, while it does a nice job of finding the normal linear signal, it can’t track nonlinear patterns in data. Of course, the data that contains nonlinear patterns can be mined from traditional insurance data. But it also originates from such places as the connected home, telematics, and social media—all the big data avenues that are currently having their day in the sun. Telematics data in particular throws traditional models for a loop. The volume and velocity of data from telematics, plus the lack of historical knowledge about what in that data is predictive are the things that make telematics a perfect candidate for a machine learning approach. Conversely, a GLM wouldn’t know what to do with it—unless the modeler knew what variables and combinations of variables to use.
So, as these new data sets become available, insurers must use advanced analytics models, which allow them to explore this data. Advanced analytics models can collect and analyze all types of data, thanks to their foundation in machine learning. One could say that machine learning is the main ingredient that sets advanced predictive analytics models apart from the traditional models.
The Importance of Existing Data, and Making It All Consumable for the Business User
It’s tempting to dive into all the new sources of data out there from the Internet of Things, telematics, and social media in all its forms. But, as we’ve already alluded to, collecting and mining it for predictive intelligence using only traditional models will prove expensive and time consuming. Before chasing down the new data (that you won’t even be able to use effectively without machine learning anyhow), commit to leveraging the data you already have. Indeed, this existing data is a goldmine because it already contains older data that paints a good picture of the future. Use advanced data mining and predictive analytics to get the most value out of the data you’ve heretofore been gathering for years before tackling the new big data frontier.
Becoming a data- and analytics-driven insurance company requires much more than data and really good models. It also requires tools that the business user can understand and leverage easily so that he or she can take full advantage of the data. If an insurer’s claims adjusters, underwriters, and sales managers don’t have a way to interact with the output of the predictive model, the initiative to adapt and succeed fails.
Of course, adapting and succeeding cannot occur without the infrastructure and technology environment to create, manage, and control multiple predictive analytics models. This is the final piece of the puzzle. Having a technology in place that deploys and manages the integration of the insights gained from predictive analytics in a low IT-touch way is very valuable.
Is Your Organization Ready?
Insurers are most successful with their predictive analytics models when the proper predictive models, data and infrastructure converge. But mindset is also key: Adapting and succeeding requires an organization to be primed to use models effectively at the business level. These insurers embrace advanced analytics, see the value in their existing data and are committed to the ongoing advanced analytics journey.
When planned for and used effectively, predictive analytics is like having a crystal ball. And who couldn’t use a tool like that?