Digging for data treasure

The Massachusetts school system knew it had valuable information locked

up in the volumes of student data it collected, but there was no obvious

way to get at it. Only when school officials employed a business intelligence

process — a combination of data warehousing and data mining techniques —

did they find a way to exploit the data.

Now, by analyzing standardized test data, school officials can create

profiles of students at risk and make early intervention plans. They can

see where to improve curricula and teaching methods.

"This approach is fairly new to schools," said John Judge, co-manager

of the Schools System Data Consortium centered in the Merrimack Education

Center (MEC) in Chelmsford, Mass., which was created to use technological

resources to address educational needs. "Educators have traditionally done

this kind of thing mostly through intuition, which sometimes works but often

doesn't."

The Massachusetts program, which started several years ago as a pilot

involving a small number of school districts, has expanded to more than

50 districts, with more on the way. The plan now is to apply the experience

gained by analyzing test scores to other kinds of data, Judge said.

Industry has used business intelligence for years to help executives

make decisions. It has shown companies how to improve customer service,

schedule tasks such as maintenance more effectively, boost sales through

better marketing and reconfigure e-commerce strategies.

It has had such a positive impact on the bottom line that it has become

a strategic necessity, particularly in larger organizations that have the

internal expertise to make use of the complex process. Because software

tools are becoming easier to use and more widely available through Web browsers,

the business intelligence market is opening up to even small and midsize

companies.

Without profit to motivate them, governments have been slower to catch

on. The Massachusetts schools program is still a rare example.

One area, however, in which officials have made more extensive use of

such approaches is health system fraud, where the payback of applying business

intelligence is more immediate and obvious. That was the case in Georgia,

where the state's Department of Audits had been able to examine cases of

fraud only after receiving a tip from a disgruntled employee or customer.

And even then, it was usually able to investigate only a specific health

care provider.

"We decided we wanted to take a more aggressive stance and extend this

to a more general examination of potential fraud and abuse," said Mike Johnson,

the department's audit manager. "With health care costs skyrocketing, we

felt we needed to do that to try and actively discourage fraud and abuse."

In fact, the department has found it can do even more with business

intelligence, Johnson said.

Just by applying the standard rules-based approach — where data is compared

to a set of rules that describe how things should work — the department

has been able to isolate many instances of violations. That does not necessarily

mean that fraud is taking place, but it gives the Department of Audits a

good starting point from which to narrow its focus.

The department is looking into expanding the process into neural net

analysis, which goes beyond rules-based comparisons of data to allow a more

free-form search for differences in patterns within the data. It's the kind

of investigation that will reveal anomalies caused by events that can't

be described by known rules.

Business intelligence is helping state health departments save money

in other ways, too. For example, Wisconsin's Medicaid Evaluation and Decision

Support System (MEDS), developed by the Department of Health and Family

Services, was instrumental in the launching of a program to provide supplemental

Medicaid for the working poor. Overall, the Wisconsin Medicaid program expects

to save between $10 million and $20 million each year through funds recovered

thanks to MEDS' analytical prowess.

Although there are many different faces to how business intelligence

is implemented, the underlying methods are the same for all. Data is first

collected in databases and segmented into groups of records that share certain

properties — they are defined by fields that contain similar values or share

broader attributes such as similar types of customers, products or behaviors.

A rules-based approach is then applied to discover associations between

those groups and to determine any patterns in how those associations occur.

A model is then built that attempts to describe the typical behavior

of a customer or user of a system. Deviations from the norm described by

this model, and their significance, are examined using statistical analysis

and visualization techniques. A variation on this uses neural network technology,

an outgrowth of artificial intelligence techniques, to reveal those associations

and patterns in the data that don't conform to known rules.

A major change in the last three years or so has been in how data mining

and business intelligence are applied, said Ben Plummer, vice president

of customer operations at Cognos Corp. Earlier tools were complex, and the

process was relatively long and involved. Now, he said, the tools have become

more visual and intuitive, and people can rapidly navigate through the data

before they isolate the pieces they want to apply the data mining algorithms

to.

Along with that comes an increasing demand to deploy business intelligence

via extranets, so those outside the organization can benefit. For a government

agency, it could mean providing citizens with a way to do their own analysis

of government information via the World Wide Web.

"The whole thing is morphing from the one-to-one process it originally

started as to a one-to-many approach, and now to one that's focused on many-to-many

distribution [of information]," said Woodson Martin, director of Americas

product marketing for Business Objects SA.

The process of business intelligence is becoming easier, and so is getting

started in it. Information Builders Inc., which already has a presence in

the federal market, is increasing its drive into the state and local market

with the planned November release of its i-Gov suite of application templates.

"[The software] will be focused around such things as regulations and

public/legal information, the areas that state and local governments see

the biggest current demand for," said Mike Corcoran, Information Builders'

vice president of product marketing.

One drawback might be price. Depending on the level of customization

and services needed, individual proj-ects could cost $30,000 and up. The

MEC program in Massachusetts charges school districts just $6,000 a year

to participate, but not many projects could be similarly constructed to

spread the cost.

That's the reason the Utah Bureau of Medicaid Fraud has opted not to

go with a full-fledged data mining option. It spends $20,000 a year to use

SAS Institute Inc.'s analysis software, according to Terry Allen, a programmer

with the bureau, "and that's already more than several other departments

spend on software combined."

Still, the MEC's Judge said he feels that the use of business intelligence

in government and education is "on its way.... Once people understand what

the tool can do for them," Judge said, "they are nothing less than enthusiastic."

Robinson is a freelance writer based in Portland, Ore. He can be reached

at hullite@mindspring.com.

Featured

Stay Connected

FCW Update

Sign up for our newsletter.

I agree to this site's Privacy Policy.