Clean as a Whistle

It's not difficult to find insurance carriers that use predictive analytics in one form or another, but how well they perform the function has a great deal to do with how much time and effort are spent on the data models.

"They all say they use [analytics] to a degree," says Rebecca Amoroso, vice chairman and national insurance leader for Deloitte. "On the underwriting side, I would say the lion's share [of carriers] definitely is building predictive models or already has implemented them into underwriting processes."

However, the level of success carriers achieve with analytics greatly depends on the quality of the data. As carriers have concentrated on scrubbing that data, the work appears to be paying off. "We're starting to see differences in some of the results [carriers] are generating," reports Mark Charron, principal and U.S. leader of actuarial and insurance solutions for Deloitte.

To get the internal data to the point where carriers can use it effectively in their analytics takes a lot of work, adds Charron. "[Insurers] have this great source for all of this data, but putting it into a common format so you can use it consistently is a challenge," he says. "We're seeing more effort with our clients around the internal data warehouse for that." And once carriers have scrubbed their data and brought it into their data warehouse, they then can go off and purchase external data, which Charron believes to be easier to blend into the data warehouse.

At Nationwide's office of investments, the data warehouse contains information from the carrier's investment accounting systems. The data has been used in a variety of analytics packages for reports and decision-making. The problem, though, was the data wasn't well documented when it was entered into the data warehouse, and it was really painful for the data team to make any changes, according to Adam Ziegler, senior consultant for solutions analysis with Nationwide.

Another issue was a lack of transparency in how data requests evolved, explains Ziegler. Projects often began with one purpose but grew into something unforeseen, creating a bottleneck. Nationwide decided to build a data warehouse and find ways to enrich the available data and combine data from multiple sources to get the best data available. "We didn't plan the Common Information Model [Nationwide's name for the data model], but it ended up driving the effort," he says.

The data model in its current incarnation is in an Excel spreadsheet that is a catalog of all the data Nationwide's investment office cares about. Ziegler claims it grew from about 300 fields in its initial implementation to well upward of 1,000 now. "What's unique about it is how central it came to be with everything we've done with this project," he says.

The first problem Nationwide's developers had to solve was the fact they were replacing everything all at once. "We had all these source systems and subject-matter experts associated with the accounting systems plus we get data from external vendors, and we had various reports and packages internally that consumed the data," says Ziegler. He went to the people who consumed the data and asked them what data they wanted, and invariably they replied, "What data do you have?" he recalls.

"One of the lessons we learned is what doesn't work is getting a bunch of people in the room who are subject-matter experts and asking them what they need because you end up with a laundry list of things that might be nice but are not going to get the immediate project done," he says.

Eventually, Ziegler and his people determined what data was needed by looking at the old system. But since the system wasn't documented, they spent most of their time defining the data. They found getting clarity about the data was one of the hardest activities in the project. "Once we had that, we consolidated," he says. "We would take data that seemed like it was similar and try to collapse the requirements into a single catalog that made sense, so we could get better use from it."

Confusion over data needs and requirements is common in such projects, according to Amoroso. There are programs and tools to help cleanse the data, but she believes carriers first must find someone internally who will be responsible for understanding the information the company has and the process for designing how the company might use that data in the future–what she calls a "data czar."

Also, it is imperative to get the business side involved early so the project is not just a technology exercise but contains some business value. "Have a plan to look continually to improve the quality and integrity of your data," she says. "Being mindful of the process and having a procedure is essential."

The data needs to be broken down into subject areas at a level where the business can understand it, Ziegler adds. "We standardized everything around those subject areas," he says. "We were able to divide up the work according to those standardized chunks."

There is one module that loads the data into the data warehouse, and another that gets data from one source to another. "Because of the way [the data] was organized using this Common Information Model, we were able to retain a lot of value that would have been lost when people walk out the door or when scattered across lots of other documents," says Ziegler. "Today, for people who are on this [data warehouse] project, the first thing they have to learn is the Common Information Model. You refer to it multiple times a day if you are working with data."

Incorporating best practices into the use of analytics requires a range of activities, explains Amoroso, including how carriers integrate their data within the organization, the training of business users, what information is communicated into the field and to inside users, and what business applications are going to be done differently once the data model is in place. Some companies have models in place, she points out, but may not be doing anything constructive with the information. On the opposite end, some companies use their data proactively and aggressively. "Those companies use analytics to better understand their customer; come up with more refined pricing; change the way they price and service accounts; and study which customers are profitable, which are not, and why," says Amoroso.

At The Hartford Life, Rachel Alt-Simmons sees her job as educating the business units on the opportunities derived from analytics. The business intelligence group in U.S. wealth management, which she directs, works with leaders from the business units. "The great part about that is when we talk business, we can talk in business terms," she says. Since the business intelligence team doesn't sit in any particular line of business, Alt-Simmons believes this brings the BI team to projects without any particular agenda. "My agenda is to work with [business units] to help develop business goals," she says. "If I know one of your business goals is to sell more this year, we'll look to see how analytics can help you do that."

On the claims side, the benefits achieved by analytics involve closer integration between the models and the business rules, explains Charron. "Models have the ability through scoring to do predictions," he says. "That's fine, but you also need to have business rules in place to change workflow to get the right claim in the right adjuster's hands."

If it's a no-touch claim situation without the need for human intervention, Charron believes carriers can increase the efficiency of the claims operation through advanced business rules powered by analytics. However, if a claim has the propensity to be litigated or involves severe medical claims or extended disability, it should be placed in the hands of a seasoned claims adjuster right from the start. "You need to start a triage of that claim early in the life of the claim as opposed to not understanding [whether the claim] is going into litigation or involves a medical catastrophe until six or nine months down the line when the opportunities to affect that claim outcome favorably have been reduced," he says.

On the underwriting side, analysts can determine variables that assist underwriters in their decision-making. Insurance vendor Fair Isaac did all the analysis for South Carolina Farm Bureau Mutual (SCFBM) in the past, using the carrier's internal data to come up with indicators that would improve underwriting decisions, according to Bruce Mackay, vice president of operations. "We started out with about 130 different variables and eventually got it down to a dozen or so we felt were highly predictive," he says. Earlier this year, the carrier purchased Fair Isaac's Model Builder software, and now the carrier does its own analysis. "We did all the data prep for [the vendor], and [Fair Isaac] crunched the data through the software," says Mackay, explaining how the system operated in the past. "We bought the rights to the software, and our project this year is to do our own analysis."

The premium and loss data the carrier sent to Fair Isaac in the past still resides with SCFBM, but the emphasis has changed to performing the analysis in-house. "It's a matter of learning how to bring the data in, structuring the fields, and doing the analysis," says Mackay.

Alt-Simmons describes The Hartford Life as a fairly conservative organization. "We tend to put more money where money is generated," she says. So, the company has invested in more analytics for its individual annuities line, which she describes as the company's "bread and butter." The individual annuities area, for example, has the best infrastructure and data, which is easy to get at, she explains. But other lines of business still sit on old mainframes, so working with them requires the BI department to make data requests of the IT department and hope IT interprets the requests correctly. "Because we're not in a line of business, we sometimes lack expertise in certain data, so there is room for [mis]interpretation," she notes. "We have to work closely with a business partner and with the IT department."

The success of any project depends on the quantity and quality of the data, continues Alt-Simmons. "The more data you have, the more likely you are to have some modelers and analysts who want to do some creative things with it," she says. "You may not have that luxury when you're working with other lines."

The Hartford's property/casualty company has a sophisticated analytic infrastructure, according to Alt-Simmons, but it also has more control over its customers. Because of different distribution channels, The Hartford Life doesn't own its customers the way the P&C company does. This has generated discussions within the life company regarding why the company would want to implement a tool to predict customer behavior when the company can't control that behavior. "The challenge has been to find the areas in which we can implement analytics where it is going to be actionable," she says. "If we are working with our wholesaling unit on marketing campaigns with brokers who sell our product, that's a direct customer. But where we've developed a retention model, where we're trying to influence clients, the brokers can perceive that as interfering with the broker/client relationship, so you have to be very cautious."

Depending on whom you talk to within The Hartford Life, determining who the customer is can elicit different responses. "Service centers are going to say everybody is our customer and we treat them all equally," says Alt-Simmons. "The sales organization will say the broker is the Holy Grail–the one it wants to make happy."

One part of the job for the BI team is to find some business value with the available capabilities. The first use of the carrier's analytics tool purchased from SAS was to build retention models in the individual annuity area for the call center. "We predicted who was most likely to buy, and based on the score, the calls were routed to a retention specialist who would have a conversation with the customer," she says.

The next opportunity The Hartford Life faced was with the wholesaling division. Alt-Simmons went there and explained a program had been developed that could target brokers who had stopped selling The Hartford Life products. "We were able to implement a number of marketing campaigns that generated multimillion-dollar lifts in sales," she says.

A third opportunity was to work with the company's office in Japan, which has a big annuity market. "We worked on the product side with actuaries on what makes people most likely to liquidate," continues Alt-Simmons. "We didn't necessarily use that as a production model but to show some triggering events and how we can build our products differently."

In working with Fair Isaac, SCFBM has been able to identify variables that are more predictive than others. This has allowed the carrier to change the application and the way customers are questioned. Over time, Fair Isaac discovered some variables are more meaningful than others. "It's an ever-changing system," says Mackay. "Of the initial 12 variables we came up with 15 years ago, we're still using 10."

Building a model to show the value of data increases investments in the data, remarks Alt-Simmons, noting The Hartford Life has taken an iterative approach in making bigger investments. "You may get new content and purchase external data that helps you build the business case further down the road to make [even] more strategic data investments," she says.

Life and annuity carriers are pitching themselves to the public as retirement entities, so that requires knowing a lot more about the situation of the customer, she comments. "Whether you can get that information from the broker and sell it as a value-add program is a big question," she says. "We don't own those brokers, so are you just giving away free leads or encouraging them to do more business with The Hartford and build more broker loyalty?"

SCFBM will expand analytics to homeowners' coverage. "The big push in the industry is moving toward using analytics to do multiple valuations with multiple variables that can be tied in together," says Mackay. "The model line looking at individual variables is kind of going away. Multiple regression techniques are seen as the future in pricing."

Mackay indicates this approach makes sense. The better a carrier can make the rate match the risk, the fairer the end rate. Many carriers use a system involving average rating, but he asks, "What's in an average? You have some people above average and some below average, but everyone gets treated the same. The finer you can slice [the data] and match the risk to the rate, you are going to have a rate that is fairer to the individual."

Having a good data model is crucial, Ziegler emphasizes, because information is a strategic capability for most insurers. "Where [good data] definitely has an impact is the speed with which we can answer questions such as: Can we do this project? How long will it take to do this project? In the original data warehouse, it was difficult to answer those questions," he says.

But Amoroso contends insurers should not be daunted by the fact their data is not 100 percent accurate. "What we have found is there still is some business intelligence you can glean from that data," she says. "As imperfect as it might be, a lot of times we find it absolutely usable. You always could improve on it, but it does bring business value." When carriers use analytics, they are looking for patterns and signals rather than precise accounting numbers, she adds. "A lot of that can be gotten from 80 percent accuracy."

But the real power of data comes when a carrier takes its internal data–gathered from all the different areas of the organization–and combines it with data purchased externally, whether it be from commercial entities or government data. "When you combine that information, you really have something," asserts Amoroso. "What you do with that knowledge is the key to differentiation in the marketplace."