Data mining is the land of what if in insurance: What if we tried to target policies to residents of this community? What if we eliminated this line of business? Most business analysts would have blanched at answering those questions a mere few years ago, but those with data mining tools and the imagination to understand its capabilities are finding the information they need to make wise decisions.
Overall, I dont think insurance carriers leverage their data as well as they can by utilizing data mining, says Craig Lowenthal, vice president and CIO of Hartford Financial Products, a unit of The Hartford. They can look at customer trends closer than they currently do and determine what product lines or coverages to get in or out of. Or even the type of industries they want to insure.
Many companies use data mining as a marketing tool. Mark Kreyenhagen, director of database marketing for Western-Southern Life in Cincinnati, believes it has changed the focus of the insurance home office. We werent a very strong marketing arm for our sales reps to use, Kreyenhagen says. The home office processed policies and paid claims. It hadnt figured out what to do with all the data. The first step taken by Western-Southern was to begin aggregating the data sales reps were collecting before those reps walked out the door. We needed to keep that in-house so when the reps left and the next person started, we could just transfer that information to the next person, he says.
Like other insurers, Western-Southern loses many rookie sales reps. When they walked away from the job, the information those reps had accumulated on their customers typically walked with them. When the data warehouse project was approved here back in 1998, one of the major benefits to establishing it was this collective knowledge wouldnt leave when the sales reps left, says Kreyenhagen.
Our companys one-year retention rate [for agents] is about 40 percent, he adds. If you are losing 60 percent of your sales reps after one year, any effort they put in before they left was typically gone. All that knowledge left with them.
Lowenthal agrees data mining is an excellent marketing tool. Thats been the general use for it, and by all means it should be used for that purpose, he says. You use the wealth of data you have and turn that data into information. Thats the whole goal of data mining.
Five Things to Like
Brian Bachman, senior statistical consultant for insurers in the data mining lab for Teradata, a division of NCR, says data mining can provide insurers five key services: scores, predictions, descriptions, profiles, and business drivers. Scores indicate the likelihood of whether a customer will purchase a product. Predictions estimate how much a customer will spend in the coming year. Descriptions define the characteristics of profitable customers. Profiles give the common characteristics of customer segments. Business drivers are the key characteristics that identify fraudulent behavior (see Clean Sweep, Tech Decisions, September 2002.)
Lowenthal agrees that fraud detection piqued the interest of many insurers. There are definitely some companies that do data mining well, he says. They use it for fraud detection, and that gets a higher profile. Fraud departments are in there looking at all the data. Theyve reduced fraud and caught some criminals.
But there are other uses as well. Lowenthal believes the biggest limits lie in the imagination of the users. There are other capabilities, he says. Why dont the line-of-business managers look at the data and really analyze what they have. In the same way the anti-fraud folks are looking at it, the businesspeople can do the same things. What type of business should you write? Whats profitable and whats not? You can look at communities within a town and find out if one is more profitable than another.
Data can tell an insurer about customers, ex-customers, and potential customers. We can take historical data on former customers and the attributes of those customers to try to figure out which customer assigned to which agent is most likely to leave the company, Bachman says. We take the data we know about the customers who left and compare it to the customers who didnt leave, and we come up with certain weighted equations to make that prediction.
There are different ways to deliver analytical intelligence to company touch points, Bachman says, but the important thing is agents and call centers have the ability to reach out and act on the information presented them. We try to synchronize insurance retention models with renewal notices so we can make a clock start ticking X number of days before a customer has to make a decision on a policy renewal, Bachman says. Once the equation has been established, the carrier can go back to its database and score the equation. This allows agents and the carrier to know which of its customers are at risk for renewal, according to Bachman.
Lowenthal believes data mining has achieved only limited success in insurance. He says IT departments have driven data mining projects, and that has to change. Business needs to understand the capabilities that are there, he notes. Business leaders have to do more than study traditional production reports. They have to try to look for other ways to use a companys vast data stores of information.
Senior management needs to incorporate data mining variables and techniques into the way it strategically plans its business, even looking at technical issues, Lowenthal adds. Senior managers still rely on tried-and-true financial reports they get monthly to manage their business.
Ask the Right Questions
Carriers should avoid generic industry equations in favor of their own models, Bachman says: We look for certain kinds of data that are commonly held. Well go into the company and construct a model for it based on data we know should be readily available, and if the company has other data that is interestingsuch as what its competitors are doingwell try to enter that into the analysis.
Adding third-party data can be valuable, though. Western-Southern purchases 50-plus data elements from data provider Acxiom (www.acxiom.com), according to Kreyenhagen. Once a year the entire file gets updated with external data, he says. If a relationship has a change in the course of the year, we send that immediately to Acxiom. Also, if there are any new relationships established, we send Acxiom a report and get an update monthly. Agents also have the ability to update the data, so we feel we have very current data.
Western-Southern licenses data from Acxiom on consumers. The insurer sends Acxiom its files, and Acxiom appends the file with customer-specific data such as age, income, number of children, home value, and home ownership. Part of that data we dont normally collect in the process of an application, says Kreyenhagen. Some of that data we do collect, but in the insurance business you can write a policy today and you may not get any substantial updates for the next 15 to 20 years. Its a unique product line. The fact that you collected application data 20 years ago doesnt help you market to that person today. With the external data, we feel we have the best of both worlds. If we write some new business, well take the actual data from the customer, but if the agent is unable to get it for some reason, were able to take the external data and fill in where were missing or where its become stale.
Profiling Potential Customers
Examining customer trends is much easier with the appropriate tools and the right set of questions. Data mining allows you to slice and dice your data a lot finer than throwing the proverbial baby out with the bath water, Lowenthal says. If a product line is not performing well, the tendency might be to stop writing that line. Well, maybe its not a bad line to write, but rather your concentration on the locations youre in or the types of businesses you are writing [should be reevaluated]. Thats the area I dont think carriers explore as much as they can or should.
When an insurer looks to attract new customers, data can be mined to create a profile of one of its customers and the profile will then be compared to lists of names with demographics from certain zip codes. Western-Southern also assigns levels to its customers (gold, silver, and bronze), so that efforts can be targeted at the top levels. In cases where we want to give our agents leads, we go to Acxiom and say, Find 10,000 leads in this zip code that match our gold and silver profile, Kreyenhagen says. Acxiom has those models already built into the system and only returns names that are clones. If we stopped there, that would be a huge improvement over what was done in the past.
The carrier can also run the names through a credit bureau to scrub anyone with poor credit. So what the agent ends up with is a list of potential customers who match the companys best existing customers or young people starting out with no credit problems who fall within the carriers target income range. External data is much more powerful for pure prospecting purposes than for retaining current customers, agrees Bachman.
Teradata uses scoring for customer records, and those scores are attached to the customers record in the data warehouse. Such scoring can be used to judge whether a current customer is a good prospect for cross-selling or up-selling. Once a carrier has the scores, it can start looking at predictive probability or other indicators that the person might be ready for more insurance or a change in the type of insurance, Bachman says. If you combine those kinds of models with other kinds of analysis, you can be more proactive.
The Perfect Vehicle
The data modeling and data mining techniques used to solve an insurers problems are largely the same, whether the carrier is looking at customer retention, customer growth, or new business, according to Bachman. No matter what you do with the data, things are similar, he says. The differences come when structuring the data set. Which data will prove more relevant? he asks. It depends on the nature of the problem and the company strategy. The toolbox is still relatively generic. Youll still be estimating the same kind of things, still applying the data, and still shipping it out to the touch points. The domain within the industry shifts from the pure retention problem to the prospecting and up-sell, cross-sell. All the analyses are integrated in the database.
The insurance industry has always been strong in analytics, but Bachman points out, At the end of the day, there are problems that arent cluster analysis or simple binary outcome model problems, and you have to have a tool.
Lowenthal doesnt feel there will be another big push toward basic data mining, but there will be companies that will up the ante. Data mining and data warehousing have had their heyday already as far as visibility, so I dont think theres going to be any kind of mass that runs to it, he says. The companies that take their data mining or data warehousing to the next level are going to see a competitive advantage or better results. Better financial results come from the deeper you mine.
Data Mining: Types and Tips
Data mining starts with data management. Wayne Thompson, product manager for software developer SAS, says insurers need tools to provide access to disparate data coming from numerous sources. It doesnt matter where the data is or what kind of platform its onmainframe or UNIX box, he says. Businesses need tools to extract, transform, and load the information into a common repository. Then you can apply the data mining algorithms to it, he says.
There are two types of data mining algorithmspredictive and descriptive. Predictive modeling tools, such as neural networks, decision trees, and regression, work well when the insurer has a pre-defined response.
Data mining tools will also help an insurer determine whether customers will buy a certain product or not, or how much they will actually buy. Most of your bang for the buck comes from predictive modeling, he says.
With descriptive data mining activities, the idea is to cluster the customers into more homogeneous groups or segments. Then you profile the clusters to determine what the driving factors are, Thompson says. Is it an older group of citizens that tends to be more conservative? Do they tend to like certain kinds of activities? It helps you position the right marketing spin and the right products around that to get them to buy.
A vast amount of time goes into data preparation, Thompson says, so once the data is prepared, a carrier might want to do some intelligent sampling. You need a good representation of churners and non-churners so the data mining algorithms can actually find the signals, says Thompson. You have all these algorithms that come from the statistical and the artificial intelligence community so you can throw multiple algorithms at the data and do some tuning of the models. The amount of tuning you do is dependent on the time you have and your experience. The ideal is you throw lots of different algorithms at itsome work well in some cases, some work well in others. You generate lots of competitive models, and from that you choose the champion model that will best indicate the opportunities for your firm.
Thompson says the Internet is an excellent vehicle for the final deployment of the data mining results, but with huge data warehouses storing millions of pieces of information, mining in real time is not practical. You go through the extract, transmit, and load process, and the business analyst or statistician provides advisory counsel, he says. They come up with the predictive models, and the results are deployed to the Web so agents and others can look at their customer base or a list of prospects. Its a good vehicle for deployment because everyone has access to it.
Data Mining Software Vendors
Data Instrument Group
Mountain Valley, Calif.
408-464-4439
www.digdb.com
DecisionPoint Applications
Beaverton, Ore.
800-970-5372
www.dpapplications.com
Fair, Isaac
San Rafael, Calif.
800-999-2955
www.fairisaac.com
IBM
White Plains, N.Y.
800-426-4968
www.ibm.com
Magnify
Chicago, Ill.
312-384-7161
www.magnify.com
MicroStrategy
McLean, Va.
800-927-1868
www.microstrategy.com
Millbrook, Inc.
Center Valley, Pa.
610-797-7400
www.millbrookinc.com
MIS AG
Newark, N.J.
973-679-0724
www.mis-ag.com
NCR/Teradata
Dayton, Ohio
937-445-5000
www.teradata.com
Net2S
New York, N.Y.
212-279-6565
www.net2s.com
Net Perceptions
Edina, Minn.
800-466-0711
www.netperceptions.com
Pinpoint Solutions
Livingston, N.J.
973-716-0700
www.pinpnt.com
SAS Analytic Intelligence
Cary, N.C
919-677-8000
www.sas.com
Thazar Solutions
Kansas City, Mo.
816-760-5000
www.thazar.com
WhiteLight Systems
Palo Alto, Calif.
650-843-3000
www.whitelight.com
© Touchpoint Markets, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.