Reading the Future

Trusting your technology partner is important in any software deal. However, that trust is even more critical in predictive analytics solutions because carriers are placing their rating and underwriting future--the lifeblood of the company--into the hands of a third party. Insurers must have faith in their vendor and need to connect with their partner on a cultural level, according to Karen Pauli, research director in the insurance practice at TowerGroup. "Does the vendor really understand what you want to be doing?" she asks. "You need to look for someone with prior experience in the same market segment you are in and someone to help walk you through the process and give you some business experience."

Island Insurance Companies is just getting started in the process of predictive analytics, relates Jeff Fabry, vice president and CIO. The carrier has been working on a personal auto project with its vendor, Fair Isaac. The situation Island finds itself in is it has three different companies that write personal auto coverage in Hawaii, and each one has a single rate, explains Fabry. When competing with national carriers, Island finds the competition has a much wider range of rating possibilities. "Right now, we have only three slots, and we're trying to increase that to six or 10 slots so people who have worse loss ratios have a little higher rate and the people with better loss ratios get a better rate," he says. "You are able to be more competitive on the better risks."

Pinnacol Assurance performed extensive due diligence before selecting Valen Technologies to perform its predictive analytics. One of the steps in that process involved sharing confidential data with the vendor to see what the technology and modeling capabilities could do for Pinnacol from a business perspective. "That was crucial for us to test drive predictive analytics because it was all new to us," says Mark Isakson, associate vice president and pricing committee chair for Pinnacol. He admits there were difficult moments in the process. "At the end of the day, because it's so new, how do you get your arms around [analytics] and absorb it before you jump all the way in? It can be a scary thing, especially for an insurance company," he says, adding Pinnacol saw the possibilities during the proof-of-concept process.

Building the models was where Isakson believes the relationship with Valen was most important because the development leveraged Pinnacol's expertise in workers' compensation. Since the vendor was building a new application, it needed to work with the carrier's existing applications and processes. "It was a fairly extensive engagement--about nine months," recalls Isakson. "We provided Valen with our data, and it would go back, do the analysis and modeling, and show us what it had. We kept refining the approach as Valen was doing the analysis, and once we had what we thought was a pretty good model--something that would work with our business model--we sat down and basically went through a testing and validating process."

Pinnacol ran the model back in time against data from prior years and evaluated how it performed against what the carrier did with its older approach to assess what the performance of the model would be like in a simulated production environment. Even more important, according to Isakson, the carrier wanted to see what some of the challenges would be in the implementation phase that were going to change dramatically the way Pinnacol did business. "We needed to go through the change management process so we could have communication strategies built and establish expectations, particularly with our underwriters, who would be using the model to price," says Isakson.

The change management piece is crucial when introducing a new pricing model and the underwriters are accustomed to what they had been doing for years, Isakson believes. "We would learn only by seeing it in a production environment rather than learning it in a live environment," he says.

The pleasant payback of the whole project was the model led to more predictable outcomes for Pinnacol because the carrier had such a huge amount of involvement in customizing the model to its business needs, culture, and philosophy. "Most of the things we learned in the process were things we anticipated along the way," says Isakson. "I don't know that would have been the case if you purchased a box product off the shelf."

Another factor in selecting the right partner is finding one that will give you the tools to allow you to run strategies, notes Pauli. Carriers need to avoid coming up with a model that will validate old assumptions. Pauli points out insurers need to see what the data tells them and then use the analytics tool to run strategies and discover the outcome. "Not all vendors have the tools to run strategies, so that's a real key for any company to look at--can you do simulations of strategies so you know what the result will be?" she asks.

The partnership between Island and Fair Isaac is vital because the vendor is doing most of the modeling. "It is important to choose a vendor with a lot of experience and that has worked with other carriers," says Fabry. "We knew who the big players were, so it wasn't that difficult a decision. For us, it came down to cost-effectiveness and things like that. Fair Isaac has a pretty good track record."

Carriers need to have a marketing and an underwriting objective in mind when analyzing their data, advises Pauli. In her prior insurance company experience, the first time Pauli ever did a model, the data suggested to her company that people who chose higher limits and added endorsements or additional coverages to their policies were the carrier's most profitable customers. "We wouldn't have necessarily known that, but that's what bubbled up from the model," she says. "To some degree, the data will suggest things to you, but you need to have an objective in mind as to what you want the ultimate outcome to be."

Pauli has found discussions between carriers and state regulators have become common for insurers entering the world of predictive analytics. Regulators find what she calls "black box modeling" to be distrustful. "[Regulators] can't always tell what the carriers are doing in their underwriting. What the regulators are suspicious of is underwriting decisions that discriminate against people of lower income or on race," she says. "If you are going to come up with an underwriting or pricing model, it doesn't hurt to work with the regulator so it can see what you are doing."

Currently, Island and Fair Isaac are going through the carrier's last five years of data and are looking for different combinations of metrics. This is where state regulations can complicate things. Fabry indicates in personal auto underwriting, age of driver is a big factor in many states, but in Hawaii, where Island does its business, that information can't be used to determine rates. "So, we have to look at other ways to determine the proper loss ratios," says Fabry. "We're looking at other types of metrics."

Fair Isaac pores through Island's data and examines the loss histories for different metrics. "Some pan out, and some don't," says Fabry. "Is it location? Is it how many miles you drive to work? A lot of companies on the mainland use credit rating, but we're prohibited from using that here. It's the same philosophy; we just have to look at it in a different way."

Island has determined it will not push the envelope with its state regulators. "We're not going to do anything that hasn't already been done here in Hawaii," remarks Fabry. "We pulled the state filings from [two competitors], and we're not going to do anything other people aren't doing in the state so we don't run into any issues with regulators."

Analytics has been around the insurance industry for a while, but Isakson believes it is fairly new to workers' compensation, which, he claims, is highly regulated in each state.

One of Pinnacol's major concerns was, with the significant investment of resources, whether the models would be palatable to regulatory agents in Colorado, where the carrier operates. "We wanted to see where the state stood on predictive modeling to see what was acceptable and what was not acceptable," he says. "A huge component to success or failure is having a conversation with your regulators. The regulators may not be comfortable in dealing with [modeling], and your investment [in the solution] already has begun."

Pinnacol explained to the regulators what the power of predictive analytics could do in its business and where the carrier saw the value. Isakson reports there were two major concerns to the insurance department. "We needed to avoid factors in the predictive model on the pricing side that were either redundant with other pricing elements already in place or discriminatory factors that might be used--such as redlining," he says.

Predictive analytics is not a once-and-done solution for insurers. "They really do have to stay on top of it," states Pauli. The downfall for some carriers is to put the model in place and then infrequently refresh the data in the model. "That's a mistake because the advantage to modeling is to keep refreshing the data so business analysts can fine-tune pricing and underwriting and not do something cataclysmic," she says.

Depending on the size of the carrier, that refreshing could be once a week or even once a month. If it's a big-volume carrier, the refreshing of the data literally could take place on a daily basis, continues Pauli. "Doing that much refreshing would call for giving the business analysts a dashboard to look into the data to see when they need to start adjusting underwriting or rating based on changes in the marketplace," she says.

The models don't have to change as often, adds Pauli, but carriers ought to revisit them yearly or every 18 months, depending on the number of transactions. "If the models are working and you are transforming your business, somehow you need to see what that transformation is doing to your results and where you need to fine-tune," she says.

Fair Isaac goes through Island's data and develops a model, the carrier implements the model into its rating, and then both teams have to stay on top of the data to maintain that model. "You have to run reports constantly," says Fabry. "Are our results indicative of what we predicted they were going to be? Is this really working out the way we thought it would? If not, we may have to do some tweaks here and there, but that really requires a lot of data analysis."

In the past, Island relied on outside actuaries for that analysis, but the carrier recently hired an in-house actuary, who is going to be taking over that job. "He'll interface with Fair Isaac, do the data analysis, crunch the numbers, and work with Fair Isaac to tweak the model," says Fabry.

It is difficult to find business analysts who have the skills to come up with new business strategies and the ability to look at the information and see what it says, according to Pauli. "It's a fairly new area," she says. "The business analysts of the future will have to be half underwriter and half actuary. Some organizations have done a good job with that, and some are still struggling. Others are using the services of the vendor they are working with, and that could be a good choice for a carrier."

Part of Pinnacol's plans over the next several years is an annual refresh of the model. "It's an ongoing process of monitoring and using Valen's abilities--Valen is the one with the technology expertise--and we're relying on it to stay in front of trends," says Isakson. "We also internally will go back through the entire validation process now that we've had six to eight months of experience with the model. We're going to do a more thorough review before setting up our pricing for 2009, and if there are changes in our underlying business we may see, we need to react--hopefully more proactively."