Get Smart

Insurers need to look beyond the typical analysis of data for ratemaking and underwriting support. The successful companies today focus on the customer and the value the customer brings to the company as opposed to focusing on products, maintains Keith Gile, principal analyst, at Forrester Research. The goal really is to find out more about the customer, to include that [information] in more of the decision-making process, and then to determine what actions to take.

Becoming customer driven means knowing who your customers are and identifying all their individual relationships with you, which is why a consolidated customer view has been at the heart of many insurers data-driven BI projects. Giles colleague, Lou Agosta, principal analyst, Forrester Research, adds carriers that write via independent agents should combine customer information and analysis with the results of monitoring agent performance in order to provide more productive agents with sales leads. Ultimately, insurers should be looking not just to understand their business but to drive the market. (See graphic, Data Warehousing Maturity, p. 26.)

Consider the evolution of BI at global P&C insurer Chubb & Son, Warren, N.J. Initially, the company looked to address its objective of cross-selling to its customer base, an effort that was hindered by customer data being trapped in different back-end systems. We were limited to viewing our business from an individual business-unit perspective. We had silos of information, says Jeff Hoffman, vice president of strategic marketing at Chubb.

In 2001, the company deployed SAS as the hub for various BI applications that would address cross-selling as well as future objectives. Chubb used Infor-matica to extract, transform, and load (ETL) internal data (including both policy and claims data as well as responses to customer surveys) and third-party data sources into an SAS customer database. The first tool the carrier developed was a Cognos-based analytic system it called Pinpoint (not to be confused with Pinpoint Solutions, a BI application vendor). The system allowed Chubb to calculate a predictive score of its customers, based on both their value to the company and their likelihood to buy additional products, and then to provide targeted leads to selected independent agents.

Pinpoint yielded higher growth rates from agents involved in the program, and as a result of that success, Chubb today is expanding the SAS system beyond personal lines lead generation and into analyses in both personal and commercial lines. One initiative will be market opportunity assessments, where the company will analyze a specific market using third-party industry data and then overlay its customer profile on that analysis to determine where it most profitably can target its sales efforts. It also will use the SAS system within its market research process to uncover hidden customer and agency needs by mining survey data.

Hoffman credits the SAS system with being the technological support for these BI initiatives. Its been a transformation from an individual business-unit view of our customers and the marketplace to a more holistic customer view and an enterprise business view, he says.

2. Is BI a magnifying glass or crystal ball?

It can be both, although most carriers focus on analyzing rather than predicting. Looking across all industries, The Data Warehousing Institute found in a recent study 75 percent of users of BI tools rely on predefined reports (for historical data) or dashboard-related tools (for operational data) to report on and monitor their area of responsibility. Twenty percent use analytical toolsfrom spreadsheets to OLAP (online analytical processing) toolsto analyze what happened. Just five percent use simulation and modeling tools to predict what will happen.

The reason insurers primarily are interested in analytical vs. predictive capabilities of BI is companies historically have lacked detailed, cross-system visibility in their operations. For instance, at Edinburgh, U.K.-headquartered Standard Life, Europes largest mutual life insurer, in order to understand the composition and value of any one fund the company offered, fund managers and investment directors previously had to consolidate data manually from a host of internal transaction processing and external market valuation systems. Standard Life sought a solution that would include a data warehouse to consolidate this information as well as an OLAP engine to allow the company to run what-if scenarios quickly to manage those funds.
Standard Life implemented an Oracle- based enterprise data warehouse and Hyperions Essbase as a dedicated OLAP database to support a half-dozen internally developed analytic tools. The warehouse gives us a single version of the truth, and Essbase is the critical part [for analytics], says Alan Carrick, senior designer at Standard Life. Essbase gives us a top-down, dynamic view of our business and allows us to construct funds without restriction.

NEW Customer Service Companies Inc., Sterling, Va., the nations largest provider of extended service policies, is an example of an insurer making use of the predictive modeling capabilities of its BI platform. According to Michael Zarember, vice president and actuary at NEW, the company first began using the ProfitCube analytical suite by Pinpoint Solutions in 1998 to avoid taxing its transactional systems with analytical requests from the actuarial department and to do the multidimensional, iterative analyses possible in ProfitCubes ROLAP (relational OLAP) architecture.

As time progressed, the company looked to the system not only to help it respond to changes in business through actuarial analysis but also to help it predict its experience and respond proactively. Today, NEW is using ProfitCubes predictive modeling capabilities to drill down to the product level, Zarember explains. We want to be able to know how many refrigerators will break down in Chicago next July, he says.
The advantages of being able to forecast this accurately will go beyond ratemaking and underwriting. We want to expand geographically and proactively our service network ahead of the [claim] volume we project, adds Cliff White, executive vice president and treasurer. If we have the service providers in place [before claims occur], its less costly and provides better customer service than if we have to put them in place later to meet demand.

3. Should BI focus on strategy or tactics?

Most companies focus on strategy, particularly when their BI initiative involves a data warehousing project. Strategic [BI] demands a data warehouse because it crosses domains, Gile says. It also requires much data and the ability to look at that data over a long period of time.

Chubb, for example, sees its market-opportunity-assessment project as a strategic tool. It is designed to help focus our marketing resources on a strategic level, Hoffman notes. We want to direct our efforts both from a customer segmentation standpoint as well as from where we are [geographically] under-penetrated.

NEW, which is in the process of deploying an enterprise data warehouse, also has a clear view of how BI affects its business strategy. We want to use our past experience to shape our future experience and maximize our [business] relationships, Zarember maintains. However, NEW also uses its BI tool on a tactical level for continually monitoring and managing its book of business. Again by using the predictive modeling of Pro-fitCube, NEW can evaluate early claims experience in new insurance programs to project pro-fitability before those programs expire. Future policies in those programs then can be priced accordingly or terms of coverage amended.

We can drill down and figure out what part of the business is causing the damage, such as high frequency or severity or a policy provision that kicks in too early, explains Zarember. Catching problems when they first start has a huge impact on the underwriting results of our book, particularly when its over hundreds of millions of contracts.

4. Is a real-time warehouse necessary?

Much has been written recently about the real-time data warehouse. However, most insurers undertake once-daily updates of their warehouse66.2 percent, to be exact, vs. less than one half of one percent that reported real-time updates in an August 2004 Forrester study. Simply put, real-time data warehousing rarely makes sense in the insurance industry or, for that matter, in many other industries. If a shipment of macaroni and cheese takes a week to get from the West to East Coast, do I need to know every 10 yards where it is? asks Gile.
Right time, not real time, should be the rule: As long as that data is refreshed every half hour, every day, or whatever is in the best interest of your business to do, advises Gile. Most data warehousing doesnt need the level of locked data integrity and table synchronization real time requires. Its difficult to pull off, and its expensive, so if it isnt needed, it shouldnt be peddled off to insurers.

Standard Life, for example, finds three updates per day, coinciding with the opening of three major financial markets (U.K., Hong Kong, and Canada), are sufficient. What is important to our fund managers is to have a consistent position. When they look [at the system] in the morning, they know their holding positions and values and the benchmark they are comparing them against are in synch, says Carrick. Real-time analysis with data valued at different times is not a current requirement.

5. How much control of BI tools should business users have?

On one end of the control spectrum, business staff comes to IT for every modification made to predefined management reports. On the other, business users are given analytic tools they can work with themselves to perform iterative analyses and create new reports without any IT involvement. Most insurers are somewhere in the middle.

We are developing some platforms that would allow us to take processes we perform in the home office and expose those to other users who are closer to our field offices, Hoffman says. What were really working through is trying to determine what is the right level of resource within Chubb that should be using these tools, so we still free up folks who need to be spending less time administering and more time selling. Those are some important change management issues.

Actuaries and other number-crunching power users typically are the first staff to be involved directly with analytic applications. Three years ago, I was the only person in NEW who used the [ProfitCube] system, and now there are seven [users], Zarember says. Because its an analytical tool that benefits the company outside the actuarial function, it gradually is moving more into the mainstream operational areas of the company, which is analogous to whats going on in the BI market in general at this time.

Before an insurer makes a BI tool available to business staff, Gile cautions it should evaluate the typical complexity of tools, which may make them a difficult fit for the mainstream. Today, the BI analytic reporting tools are very good for power users, adequate for business users but not very good for them, and very poor for casual users, he asserts.

However, even though business users may not use the tools directly, they are having more of a say in what an insurer chooses. The choice [of systems] is being driven more by the business side, points out Steve Romaine, vice president and senior consultant at the Robert Frances Group. In the past, IT said, This is what the solution is going to be. [Today] there is a better collaborative environment on the front end.

6. What is the biggest data challenge?

Quality, quality, quality. Without clean data, any analysis is likely to be far off the mark. The oft-cited problem of disparate systems that capture different dimensions about the same customer (or even the same policy) is a key culprit as are the shortcuts taken by data-entry staff to get its work done.
Although insurers understand the problems of data quality, it has been a difficult issue to solve because it is part technological and part cultural. As Dal Cais Research put it in a 2004 report on data quality, Data owners refuse on principle to admit the quality of their data is anything less than 100 percent accurate.
Other times, the data is right, but the understanding of it is wrong. Life and health insurer Universal American Financial Corp. (UAFC), Rye Brook, N.Y., addressed this issue head-on when it created its analytical data warehouse with the help of BI vendor Insight Decision Solutions. The primary system we were bringing data from had been customized over the years, and a lot of the written documentation was outdated, says Dianne M. Jessie, senior financial analyst at UAFC. Sometimes the original copybooks would say there were only eight valid entries, and we would find there actually were 200 values input in that field.

The solution was what Jessie characterizes as an iterative process that started in 2002 and culminated in the deployment of the warehouse, Insights ETL tool to extract data from two primary mainframe administration systems, and the Insight Enterprise analytic suite in 2003. She reports the system could have been deployed more quickly had UAFC known the full extent of data-quality issues and assigned more staff to the project.

However, with the system now in place just over a year, UAFC has noted its impact. Its saved hours and hours of people time in getting questions answered and reports created, Jessie says. A lot of [the analysis that] had been done before involved users taking data out of databases and systems and into Excel or Access, manipulating it, and generating queries. Sometimes, information was being taken off printed reports and reentered. The [Insight] system has taken manual intervention and replaced it with automated reporting.

Ensuring data quality is an ongoing process. At Chubb, quality assurance starts with its Informatica ETL system, which monitors not only internal data but third-party data (such as publicly available census and corporate financial data) that is matched to the customer file. We receive a confidence code that tells us what percentage of customers matched, how much data actually has come along with that match, and what is the consistency of the data that is included within that match, Hoffman says. Once data is merged, another automated data quality check takes place before its loaded into the warehouse, and any exceptions are sent for manual intervention. We continually work with external vendors to improve the process, he adds.

Closely related to the challenge of data quality is the issue of data modeling. Without an accurate and comprehensive data model, an insurers warehousing project ultimately could result in a mass of data thatwhile consolidated and easier to accesslacks structure, leaving it to users to transform the data and making it difficult to analyze. But as with data quality, turf wars can make it difficult for an insurer to agree on a common model unless the effort has top-level support.

UAFC found that working with a niche vendor that had an insurance-specific data model out of the box helped facilitate the modeling process. If we didnt have a company that understood insurance working with us, the [implementation] timeline would have been longer, Jessie says.

7. What are the basics to BI project success?

The most important component to the success of a BI project is to define business objectives before selecting the tools to accomplish them. While this is hardly a uniqueor even newrecommendation, its one Gile has seen insurers ignore time and again. Insurers try to standardize on an [enterprise] BI tool, he says, when it doesnt support all the user communities or business intelligence objectives they want to target. He identifies five such communities: IT, power users, staff who use data from BI applications, staff who use business reports, and extended enterprise users, such as brokers and agents.

A successful BI project might involve selecting different BI tools for different objectives or departments, particularly as an insurer moves those tools outside the IT and power user communities, Gile suggests. They should match tools with user classes and which strata within the organization that come into play, make smaller commitments to multiple [vendors], and deal with any integration problems, he continues. Instead Ive seen companies force a single tool on users that was too powerful and complicated, required too much training, and was not being used by those it was given to. They need to put the business back in business intelligence.

Insurers also need to emphasize basic punt, pass, and kick skills when undertaking a BI project, advises Romaine. Leading companies are focused on benchmarking and best practicesidentifying key performance indicators and not just going off on a fishing expedition, he says. They scope the project and get buy-in. And they start incrementally and build on success as opposed to the waterfall approach.

Tech Guide: Business Intelligence Tools

Acorn System
Houston, Tex.
713-963-9000
www.acornsys.com

ALG Software
Atlanta, Ga.
888-374-4321
www.algsoftware.com

Allegient Systems
Wilton, Conn.
203-761-1289
www.allegientsystems.com

Allfinanz
New York, N.Y.
888-824-2929
www.allfinanz.com

Appfluent Technology
Arlington, Va.
703-284-0800
www.appfluent.com

Ascential Software
Westboro, Mass.
508-366-3888
www.ascential.com

Brio
Santa Clara, Calif.
800-879-2746
www.brio.com

Bristol Technology
Danbury, Conn.
203-798-1007
www.bristol.com

Castek
Toronto, Ont., Canada
416-777-2550
www.castek.com

Centive
Bedford, Mass.
781-778-8000
www.centive.com

CNR
Reno, Nev.
775-851-2829
www.cnrsearch.com

Cognos Inc.
Ottawa, Ont., Canada
613-738-1338
www.cognos.com

CS Stars
Amarillo, Tex.
800-858-4351
www.csstars.com

Digital Sandbox Inc.
Reston, Va.
703-390-9770
www.dsbox.com

Guidewire Software
San Mateo, Calif.
650-357-9101
www.guidewire.com

The Haley Enterprise
Sewickley, Pa.
412-741-6420
www.haley.com

HCL Technologies of America
Schaumburg, Ill.
847-303-7200
www.hcltech.com

Infoglide
Austin, Tex.
512-532-3500
www.infoglide.com

Information Builders
New York, N.Y.
212-736-4433
www.informationbuilders.com

Millbrook, Inc.
Bethlehem, Pa.
610-867-7400
www.millbrookinc.com

NCR/Teradata
Dayton, Ohio
937-445-5000
www.teradata.com

PeopleSoft
Pleasanton, Calif.
925-225-3000
www.peoplesoft.com

Pinpoint Solutions
Livingston, N.J.
973-716-0723
www.pinpnt.com

Results International Systems (RIS)
Worthington, Ohio
800-875-2126
www.resultscorp.com

SAP
Newtown Square, Pa.
610-661-1000
www.sap.com

SAS Institute, Inc.
Cary, N.C.
919-677-8000
www.sas.com

Service Integrity
Newton, Mass.
617-965-0281
www.serviceintegrity.com

Skywire Software
Frisco, Tex.
972-377-1110
www.skywiresoftware.com

Spotfire
Somerville, Mass.
617-702-1600
www.spotfire.com

Thazar
Overland Park, Kans.
913-327-7881
www.thazar.com

Universal Conversion Technologies
Addison, Tex.
214-348-2000
www.uctcorp.com

Valen Technologies
Denver, Colo.
720-570-3333
www.valentech.com

Visibillity, Inc.
Chicago, Ill.
888-484-7424
www.visibillity.com