Insight: Uncovering Data

You can call it predictive modeling, data mining or analytic technology – there is some confusion about what these terms actually mean today in the property and casualty insurance industry – but if you believe in the aphorism “follow the money,” it is clear insurance companies are investing heavily in how to collect, access and use data to segment risk, predict likely future outcomes and lower loss ratios. In short, insurers want to better tap into the massive amounts of information they have to generate profitable growth.

The general field of analytics quickly gets into the somewhat space-age terminology of neural nets, pattern recognition, machine-learning algorithms and evolutionary computation. What all these high-tech software tools or engines do is automatically examine data, identify risk factors, build models and deliver conclusions to business managers. Or at least that is the theory. And it has been shown to work in developing sophisticated rating “cells” or underwriting “tiers” for individual risks.

“The most ubiquitous application of predictive analytics is in underwriting – more specifically, in personal lines auto insurance,” says Mark Gorman, the strategic research advisor in insurance for the consulting firm TowerGroup. “That is being driven completely by market forces and a couple of major competitors in the marketplace.”

Insurance companies like Allstate, Geico and Progressive in the United States have used predictive modeling in direct auto insurance to set prices and underwrite customer segments for years. Some think this is raising the bar for the rest of the industry.

“Using predictive models to price, underwrite and market personal auto insurance is fast becoming a competitive necessity, if it hasn’t already” says Frank Coyne, chairman, president and CEO of ISO at a recent technology conference. “Visionary insurers will soon be using models based on hundreds and even thousands of variables for individual addresses. These next-generation models have already been developed, and they are much more powerful.”

THE DATA BATTLE

Data analytics has become a top issue for insurance company senior management, not just IT departments, according to several sources. James Barber, North American sales manager for Information Builders, relays a quote from the chief actuary for Hartford Insurance at a conference of the Insurance Data Management Association: “There is a data battle going on, and if you are not aware of it, you are already in it – and losing.”

Adds Gorman: “Four years ago, I was in the vendor community selling these kinds of solutions into the marketplace and we were having difficulty getting traction. Those organizations not using it today are behind.”

Some evidence suggests top insurers in Canada are using predictive modeling for pricing and segmentation. But few studies or tangible examples point to how much it is used or how widely it is adopted here. (Several phone calls to insurance companies were not returned for this article). Many cite business intelligence and proprietary company strategies as reasons for the dearth of information on the subject in Canada.

“I think another part of it, quite frankly, is embarrassment about the state of the nation,” says Barber, whose company focuses on enterprise business intelligence and data solutions. “People talk about predictive this and neural that, but then you look at the fact that many insurance companies cannot count their claims in any consistent way across their lines of business, or define what a risk is numerically. There is a huge gap between some of the fluff and the reality on the ground.”

Barber notes that among Canadian companies in the top 10% in premium size, “we are seeing some focus on underwriting segmentation and pricing segmentation because that is how (insurers) are going to impact the loss ratio,” he says. “We are aware of companies that have cells for underwriting, so it is pretty much a one-on-one underwriting. There are more companies moving in that direction.”

INDIVIDUAL RISK RATING

Doug Johnston, the vice president of interface services at Applied Systems, says we are really seeing “a move more toward individual rating of risk characteristics rather than slot rate. One of the things that companies are finding is all sorts of interesting relationships and correlations in an individual insured’s data elements that can better predict future risk.”

Citing an example, Johnston notes how a carrier might look at the correlation between when customers paid their bills (immediately, on the renewal date or late) and claims. “Carriers never really had the ability to go into their billing systems and pull that data and measure it against their claim system and say, is there predictability here?” Johnston notes. “That is what these analytic engines are doing for carriers.”

Barber says that for the primary company, the key is going to be pricing and loss ratio. “At the end of the day, you are selling an insurance contract,” he says. “The core of the insurance problem is you don’t have a cost, you only have a probability of a cost. So how well you understand the probability of that cost and segment and pool it, that is your core business. The informational answer to that has not really been seen as part of an insurer’s core business. Instead, people talk about policy systems and transaction costs. But do you really think a company like ING is profitable in Canada because they have lower transaction costs?”

Insurers have only recently been able to pull together this information from diverse sources of data and technology platforms. “If you look at it historically, carriers have had completely disparate systems – from policy to billing, claims and statistical reporting – and it was very expensive for carriers to tie those things together into a process, let alone tie in correlation,” says Johnston. “But now, with some of the new tools and technologies, it is easier to write data out to data warehouses or data analytic systems, where these relationships can be measured.”

Results are starting to emerge after an expensive trial-and-error process of creating data-marts and vertical information “silos” that could neither be shared across an organization, nor actually help business managers make decisions, according to Barber. He contends that much of the “other 90%” of the Canadian market faces an uphill battle to catch up to more sophisticated competitors that are already experimenting with data mining solutions.

“The game is for the 90% to get to the base level and start to exploit the data,” Barber notes. “They are talking about a dozen or two dozen rating cells. Well, whether you have 50 or 1 million rating cells, you are going to get into diminishing returns. As soon as you are good in one area of informational analytics, the next best return is the area you haven’t done yet. You are going to leap over into pricing, and then expense management and fraud.”

“Several insurance organizations are currently grappling with the issue of whether they have the data and expertise necessary to build it themselves or buy an off-the-shelf model from a vendor,” notes Gorman.

EXPLOITING ANALYTICS

The competitive reality for companies that don’t use data analysis tools is that they may be off on pricing, poor at segmenting profitable/unprofitable risks or adversely selected against by more knowledgeable competitors. “If they are not doing it, they run the risk of only writing the business where their premiums are the lowest because of their rating tiers,” Johnston says.

Carriers with fairly accurate predictive modeling systems will have an edge over their competitors, Johnston says, because they will be able to figure out how a risk can be rated for a lower premium. “In fact, some carriers are saying: ‘We are almost getting adversely selected against because we are not involved in predictive modeling,'” Johnston says. “That is an interesting concept.”

This competi
tive reality is expanding away from just personal lines into several other areas of insurance (such as commercial lines, for example), according to research groups like Conning and Company. In a survey released last year, Conning research director Stephan Christensen stated that “commercial insurers are inspired by the advances in predictive modeling that are revolutionizing the personal lines marketplace.”

Gorman says the “market is really heating up and becoming much more dynamic” in commercial lines. “It started in small commercial lines, but it has already moved to medium and larger risks,” he adds.

Reinsurance companies have also shown some interest, particularly in how predictive modeling can help in the areas of loss reserving and incurred but not reported (IBNR) claims. Barber notes that Folksamerica Re worked with Information Builders (and another actuarial application provider) on an enterprise data warehouse to develop a fully automated loss reserving system.

There seems to be an increasing appetite for these data tools in many facets of the insurance process, according to Gorman. “I am seeing a much broader application of predictive analytics than just risk and underwriting,” he notes. “The issue is all-around enhanced decision making. So for underwriting, I want to predict my loss costs. From a claims standpoint, I may want to be able to predict subrogation or fraud or my reserve requirements. From a marketing standpoint, I may want to be able to predict retention, conversion, product preference or customer profitability. It is really about analyzing data to predict future outcomes.”

There is also an increasing emphasis on so-called “executive dashboards.” These are easily visualized and intuitive data presentations on various elements of the business for senior management. “We are also seeing a big focus on executive metrics to allow companies to be agile in their business,” explains Barber. “There has been a huge shift to an enterprise approach because the executives realize they need it to compete. They realize the claims data isn’t just for claims – actuaries need it, executives need it, underwriters need it.”

Gorman says insurance companies are demonstrating a desire for additional data. “We are seeing that emerge in two areas,” he says. “One, they are looking for better data from data sources. For example, with Progressive putting black boxes in vehicles, they are using these to track not just how far you drive, but when you drive and on what routes. That is totally different data. The second area is ‘derived data,’ meaning data created from two data elements that currently exist. We will see much more of this analysis, especially in marketing campaigns.”

EMERGING CHALLENGES

As the insurance industry moves forward with data analytics, challenges are emerging, several sources note. One of the biggest may be the temptation for insurers to merely tackle point solutions without looking at how data needs to be collected and accessed across all parts of an organization, according to Barber. “The issue is that you don’t want to have to look at each one of these things in a silo environment,” he says. “Then you are absolutely dead. The natural way to nail this is to do an enterprise data warehouse and get the data in one place, so you can drill down and navigate. Once you have done that, you have broken the back of this.”

Another obstacle is linking the conclusions from data to the ability to act and execute business strategies in the marketplace. Glen Piller, president and CEO of iter8 Incorporated, asks: “What happens once you have this wonderful information to act upon? For many companies, a pricing change could take months or even a year. Making a product change is often a huge undertaking if it requires changing codes embedded in legacy policy management systems. We are supportive of predictive modeling, but our advice is when insurers select a Business Process Management solution, they should build in the ability to act on the data intelligence they are getting very quickly.”

Data analytics that insurance companies now use are creating significant changes to their brokers’ workflow. “Brokers may have seen this, but perhaps not understood the reasons behind it,” says Johnston. “More brokers have to go to the carrier’s Web site for quoting and submissions. The reason is that the carriers cannot fully document externally what their underwriting calculations or tiering methods are. I think we are at a crossroads here, where it is no longer practical to say you can gather data in a broker’s office and fully calculate that risk without ever leaving the broker’s office. The data now needs to be transported to the carrier where their engine is running.”

For many insurance companies, the competitive benefits of data analytics outweigh the risks and challenges that might lie ahead. Sources say the hum of predictive modeling activity lies underneath many insurance companies’ technology plans. That hum is only expected to increase in volume in the future.

“I don’t want to make it out to be a survivor-mode initiative, but the level of accuracy in (predictive modeling) is so good that almost every company I have talked to is either on their way to it or researching it right now,” says Johnston. “We ask companies what their Number 1 technology initiative is for 2007; in almost every case, it is predictive modeling. It doesn’t matter whether it is Canada or the U.S.: they are all retooling their back ends to make it happen.”

As Barber puts it: “Adjusting to changes in the market requires a good data environment underneath the business strategy, and companies are recognizing that their informational stuff needs work. Before, it was an inconvenience. Now, it is a competitive reality.”