Predictive Analytics: Four Steps to Data-Driven Success

Simply building models will not enable effective use of predictive analytics. This four-step process will help position insurers for leadership and longevity.

(Image credit: Adobe Stock.)

According to The Chartered Insurance Institute, 82 percent of insurers recently surveyed agree that analytics has the power to transform insurance organizations into true data-driven businesses.[1] However, surprisingly, another study conducted by Strategy Meets Action (SMA) found that only 38 percent of property/casualty insurers feel they are effectively using predictive analytics.[2] Why the disparity?

Simply put, the answer to this question is that effectively using predictive analytics across the entire business can be complicated. Indeed, there’s a lot to consider, such as who will use it, how to make the data usable, and how insurers can oversee the operation and impact of the models on an ongoing basis. Insurers making a successful go at predictive analytics have in place a four-step process. It begins with building great models, but it certainly doesn’t stop there.

Step 1: Build Models

Obviously, the very first step is to build models. To do this, insurers must acquire and prepare data, set it up, and choose some modeling technique, whether generalized linear models or advanced analytics models. Whichever approach is chosen, models must be built on sound methodology that enables accuracy, credibility, and repeatability for multiple future models.

Remember that part of building models is understanding how, when, and where the models must be used: What systems will they touch? What business rules must be changed? Where is the data to build the models coming from, and where is the data to use the models in operation coming from?

Many insurers mistakenly think that simply building models means they’re using predictive analytics effectively. The problem with stopping here, however, is that it doesn’t yield any kind of business result. For that, insurers must deploy them—a critical piece of the puzzle we’ll tackle next.

Step 2: Deploy Models

Deployment is arguably the biggest hurdle in the four-step process. In this phase, the models are integrated with the other systems. Take a strategic approach by starting deployment with a system that will make the output easy for the organization to consume.

Often from an IT integration perspective, the easiest place to begin deploying models is in pricing. The ratings structure is likely familiar to the company as a whole, and a new model that creates a new variable is pretty easy to plug back into a ratings structure.

However, keep in mind that it can be a bit more complicated to determine which models to deploy to underwriters or claims adjusters to inform the underwriting or claims-handling processes, for example. Asking the IT department to program complex models into the underwriting or claims workflows is a lot to ask.

Step 3: Interact

If insurers want predictive analytics to make a difference in their business, it must be deployed in a way that allows the business user to easily interact with the output. For example, if a claims adjuster or underwriter faces a sharp learning curve and must jump through an inordinate number of hoops to use the information, it won’t get used. The best way to avoid this is to embed predictive analytics directly into the business user’s workflow whenever possible.

Step 4: Monitor

Many insurers don’t think about monitoring until it’s too late. They’ve built the models, deployed them—and then realize that they must have some way to monitor these now-operational systems. Are they working? Is the data right? Even if the models are working well, are things changing? Should they be refreshed?

Let’s face it: A data-and-analytics-driven organization delivers better business results all around. Executives get a comprehensive system that uncovers measurable results, and business users get easy-to-consume information that helps them perform their jobs more effectively. Essentially, predictive analytics transforms the business and becomes fundamental to every aspect of it.

But simply building models will not enable effective use of predictive analytics. This four-step process—along with the technology that helps achieve all four steps seamlessly and in an integrated fashion with as small of an IT footprint as possible—will help position insurers for leadership and longevity.

[1] The Chartered Insurance Institute, The Big Data Rush, April 26, 2013

[2] Strategy Meets Action, Data and Analytics in P&C Insurance: 2016 and Beyond, Mark Breading, June 2016

Wade Bontrager // Wade Bontrager is Vice President, Business Owner, Predictive Analytics at Guidewire Software, and the former CEO of EagleEye Analytics, which Guidewire acquired in early 2016. Bontrager has an extensive background leading analytics driven insurance organizations, having been a senior executive at several leading insurers.

Leave a Comment

(required)