Hazardous Data

Its an easy story to sell to upper management, Fridenberg believes, especially when you can point to specific problems and potential problems that will arise from not having data-quality initiatives, checkpoints, and touchpoints. When you are dealing with 15 million customers and consider the volume of change and the number of transactions we do a day, the impact [of poor data quality] would be pretty severe, he says. Implementing data-quality initiatives can be costly, but the potential for savings is tremendous. Its not a hard sell when you can tell a story of what the potential could be by not having stringent data-control initiatives in place, says Fridenberg. I believe you could draw a good business case for ROI by putting in data-quality initiatives. The potential for creating impact is substantial when youre dealing with large systems such as ours, so the ROI is there through cost avoidance.

Farmers has worked hard on improving the quality of its ad hoc marketing data for over three years, according to Fridenberg. And the company has found poor quality has an impact on customers, agents, and service centers. This is a serious business, and we take our applications and our data very seriously, he says. The importance of data quality is ingrained in us.

Multi-line insurer Consecos decision to bring its data cleansing in-house was financial, asserts Tom Besancon. As assistant vice president of marketing information and technology, Besancon says, We turned to it to save us some moneyreduce costs in terms of standardizing and cleaning up some of our [customer] names for mailings.

The suite of tools Conseco purchased from software provider First Logic not only helped to clean up the companys data, it offered the opportunity to enhance the insurers data as well, particularly some of the prospect lists Conseco purchased for marketing its products. We were looking at the tools to improve some of our modeling efforts by using matching and consolidation to get a single view of the customer, he says. We are doing a lot of ad hoc merge purgespurging privacy mailings and previous campaigns to reduce costs and stay in compliance with the privacy stipulations and laws.

Fill in the Blanks

Cleaning up data is important, Imhoff believes, but improving data can be done easily if everyone dealing with the data understands the needs of other departments within the company. One of the problems I see all the time is the claims team will fill out forms so it can do its job, which is to process claims, she says. There are lots of fields in a claims form, though, which have nothing to do with paying for the claim. Some of those fields are useful down the road in analyzing the claimhow did it happen, what were the reasons, what were the dates of the claimand can be very useful in showing patterns of fraud.

The problem is those fields are meaningless to the people trying to close the file. They simply say, Is this a valid claim? If it is, they pay them, says Imhoff.

Administrative people entering data dont realize what they are entering into the system is being used elsewhere in the company, even if no one in that particular department is using it. Sometimes its just recognizing the problem, she says. Once its recognized you can start to say, What can we do to fix the problem?

The ROI

At that point, Imhoff suggests, a bigger problem comes into play: ROI. How do I change someones business processes without affecting the bottom line? she asks. If I make claims clerks look up codes or verify dates, I slow them down. If they are like most order-entry clerks, they are paid by the number of claims they enter in a day. Part of the quality problem is looking at the holistic view and saying, Is it worth the time its going to take and the money were going to lose because were not getting in claims as fast as we used to?

Most companies have the correct information about a customer, they just have too much information about the same person, and that leads to confusion. Imhoff says data-cleansing tools really shine in this environment. If it is the customer information youre worried about and you have multiple instances of the same customer, then the tools can handle that easily, she says. Using her own name as an example, she says Claudia Imhoff can appear on data as C. Imhoff, C.M. Imhoff, or Dr. Claudia Imhoff. All these are different versions of me, she says. Data-cleansing tools can clean all that up and consolidate it into a single record, which is what you want.

The (Almost) Perfect Data

There is no such thing as perfect data, according to Imhoff, but companies can get close, and the closer they get, the less money its going to cost them in the long run. You try to get the data to the point where it is as good as it possibly can be, she says. There are always subversive things that will cause the data to be maybe 99 percent perfect, instead of 100 percent, but thats better than what most organizations are dealing with today.

There is more than just customer information to contend with, however. Insurers have product and claims information, and health insurers have lists of providers. Are they in there multiple times, says Imhoff. You dont want to go in and fix the whole thing at once. You want to go in piece by piece and slowly work through the enterprise data.

That means establishing priorities, though. What most insurance companies do that Im familiar with is to prioritize the data and decide which pieces are most critical right now, says Imhoff.

Fridenberg describes it as peeling back the layers of an onion. The decision has to be made to start with the areas insurers feel will have the most immediate impact on their customers, the agents, and the service centers. Those things are typically anything that has to do with balancing, in the sense of accounting or GL, he says. Commissions are always top of mind, or anything fee related or premium related. Those are things that are touchpointscritical path items.

Cynthia Saccocia, senior analyst in the insurance practice for the research consultant TowerGroup, believes the uniqueness of insurance contributes to the quality of data it can collect on its customers. She believes that is one reason a number of companies are pushing to have their independent agents licensed in both the P&C field and with financial products.

Insurers have typically maintained a silo organization, she says. They are dealing with a long legacy of doing business a particular way. Now, convergence in the marketplace is pushing them into a space they havent been accustomed to working in, and they are yet to get really comfortable.

She believes data collected by insurers is superior to what financial services companies can get on their clients. Typically, a customer goes into a bank for episodic-type advice, says Saccocia. Something that is very oriented to point in time. Insurers have a wealth of data they can support in cross-selling opportunities and be very targeted.

Ive Got (Algo)rithms

The software tools are incredibly sophisticated today, Imhoff points out. A program is made up of hundreds or thousands of algorithms that comb the data. Once you set [the program] free, it is off and running and can do whatever you want it to do, she says. Its just a matter of getting to that point. Imhoff advises insurers not to try to build their own tools, though. You would have to write those hundreds if not thousands of algorithms, she says. You dont want to do that. You might as well go into the [data-cleansing] business if youre going to do that.

Insurers should consider one example, she explains: What if you wanted to send a mailer and you realized 20 percent of [the names and addresses] were dupes? she asks. How much money would you save if you didnt spend the money [for the duplicates] on the postage and so forth. I imagine [the savings] would more than pay for the tools, especially if you have millions of insurance policies. Now youre looking at a customer instead of five.

Finding the Right Tools

Insurers shouldnt complicate things from the outset, Imhoff warns. I would look for a tool, first of all, that is easy to use, easy to set up, easy to understand, and easy to maintain, she says. Dont expect all data-cleansing tools to be simple, though. Some of them can be difficult, she adds. They are very esoteric in some respects, so I look for ease of use.

The second step in the selection process is the level of sophistication required by the insurer.

A third step is determining whether the system maps to the current technology the insurer has in place. Can you use it on all your databases, or are you limited to mainframes? she asks. You need to look for a map in terms of the technology.

The fourth step often is overlooked. Thats support from the software company itself, she says. I dont think most people think about that. She believes the vendor has to be fully supportive of the project. Will they help me, not just working the tool, but in analyzing the results from what the tool gives us, she says. What is this data telling me? What do I need to do? What needs to change?

Besancon says Conseco had a specific need when it began its product search. We were looking for something that could sit on a UNIX box and process loads of data quickly, he says. When shopping around, insurers will find a wide array of products. There are products out there that are PC based and larger products as well, he says.

There also are plenty of outsourcing options, Besancon mentions. Although Conseco has purchased its own tool, it is keeping its options open. The First Logic product has a suite of tools, which will allow Conseco to add on over time. But the company also kept some of its outsourcing options open. Were not putting all our eggs in one basket, he says.

The historical problem of data for insurers deals with the expensive process of building a data warehouse, trying to cleanse the data, and defining it, according to Saccocia. Insurers have had a lot of starts and stops with those types of activities, she says. Now theyre at a point where they have to step back and say, Are we doing a good job in building this? What are our core objectives to achieve it? That is when they need to look for vendor support to move to the next level.

She believes the typical approach to data solutions for insurers was to build a system because insurers felt they had unique needs. In our conversations, were finding many vendors that may be horizontal in nature and are trying to become more vertical for the insurance industry, says Saccocia. The vendors can provide specific tools for an individual industry. The data is there, she says. It just needs to be better managed. This will become more important as insurance leaders change their view of what an insurance company is. They are becoming more brokerage oriented, and they view themselves as competitors in the financial services marketplace, says Saccocia.

Worth the Effort

Conseco is saving a great deal of money today in printing costs and postage because it has eliminated many of the duplicate names and addresses from its database. But from a marketing standpoint, Besancon believes the ability to track the success of a mailer will be invaluable. We go back to the tool for response analysis, he says. If we do a mailing of, say, 50,000 people, we actually can determine if they bought a product by checking the database. He claims the company had ways of gauging its marketing success prior to purchasing the software, but the cost was just way out of hand.

Besancon says Conseco has surpassed all expectations of the project. Our success rate has been better than imagined, he says. We didnt realize how much money we could save by bringing this in-house.

How to Launder Dirty Data

James Fridenberg, vice president of applications with Farmers Insurance Group, says the data-cleansing process starts with the management of the company. You have to get management to understand the problem, he notes. You have to get its buy-in or support, otherwise youre not going to go anywhere.

Once management is on board, a focused, dedicated initiative needs to be undertaken. It has to be staffed with very smart people focusing on defining data quality in the sense of what are your critical path itemsat least the first couple of layersand what are the things that will cause the most pain when you are starting out.

The next step is to identify the appropriate testing scenarios, scorecards, indicators, and red flags. Once you get a handle on defining those items, set up criteria to look for those and a process to solve them.

Once you've determined your processes and methodology, establish best practices and start on the next layer, Fridenberg says, adding, "Use the processes and keep driving it through."