Magnifying Data

Predictive analytics begins to prove its worth in government.

by / April 27, 2005 0
Whether it's referred to as text mining, performance measurement, predictive analytics, enterprise analytics or business intelligence, software applications that help manage and analyze data are finding their way into public-sector agencies -- sometimes with inspiring results.

Oftentimes government agencies possess the data to reach constructive conclusions, but have no way of putting the data into a usable format. That's where analytics come into play. Analytics are especially useful in areas where huge amounts of data are amassed but often not used constructively, such as billing fraud, audit selection, law enforcement and as an addition to ERP systems.

Analytics can be used to determine where crimes are likely to occur within a given time frame, as the police in Richmond, Va., demonstrated when they overhauled officer deployment on New Year's Eve.


Practical Purposes
The software differs by name from vendor to vendor, but the basic premise is the same: Convert data to streamline processes, save money and increase productivity.

SAP's platform, sometimes called a business intelligence portal, offers a data management system that organizes and analyzes data from varied systems and organizations, then presents it to users through a common portal.

"We provide full-blown enterprise software," said Steve Peck, president of SAP Public Services. "It's integrating business processes for companies of all shapes and sizes, and entities of all shapes and sizes, helping them integrate the business processes and become more efficient."

SPSS Inc.'s analytics software uses algorithms to develop "clusters," or packets, of useful information derived from statistical analysis or text mining, presenting those clusters to the user through a graphical interface.

Dallas' 311 system, though extremely useful to city residents, creates a problem for the city because of the vast amounts of data pouring through the system. To get a better handle on what the data could tell the city, CIO Brian Anderson enlisted Cognos software to develop performance measurement strategies for more effective management.

Anderson said he saw the benefits of data analysis while he was CIO of Philadelphia, where the police department spent $11 million per year on overtime. Analysis of the data revealed that a district attorney policy cost the police way too much money and manpower.

The policy called for numerous officers to be listed as arresting officers in each arrest, and all arresting officers were required to be in court with the arrestee. For every officer in court another had to be paid overtime to fill his regular shift.

"Sometimes we had 15 [officers] for every arrest sitting around the courthouse," Anderson said. "In identifying the issue, we were able to change the management practice and saved millions."

The spreadsheet numbers didn't show the trends, he said, but the analytic software did.

One key area for state and local government users of SPSS analytics tools is Medicaid fraud, according to SPSS' Public Sector Technical Director Bill Haffey. New York state contracted with SPSS to develop a data-mining or predictive analytic model that would detect fraudulent Medicaid claims.

The model included five years' worth -- more than 10 terabytes -- of data from more than 52 counties. The system uses algorithms to group providers with similar behavior. Provider groups are defined by volume of Medicaid claims, type of procedure, or demographics of their patient group. Once the software develops provider groups, Haffey said, sparsely populated groups of providers arouse suspicion because those providers are billing in unusual patterns.


Commercial Origins
The concept originated in the commercial sector. Haffey used an analogy of a grocery store manager watching shopping carts to see who buys what.

"With that kind of information," he said, "you can make marketing decisions, such as putting bananas in the same isle as cereal or maybe on the other side of the store so the shopper has to walk from one side of the store to another and picks up things along the way."

A more progressive example is a T. Rowe Price call center. Since it's known that people who buy mutual funds also tend to buy insurance policies, callers purchasing mutual funds are immediately tagged as potential insurance buyers, said Haffey, adding that putting analytic software in a commercial perspective makes it easy to see how the software could be useful in government applications as well.

The concept is put to the test in auditing. States have typically used the "volume exercise" to choose which corporations to audit by looking back at corporations audited in the past and where the payout was large. Those corporations would be targets for audit again.

Analytic software can predict, down to each corporation, what the tax adjustment would be if the corporation were audited. With those numbers, gathered from sources such as the previous year's returns, government can be more precise in targeting corporations that will yield large adjustments.


Analytics and Law Breakers
Police in Richmond, Va., plan to expand the use of analytics to boost efficiency after successfully using the software tool to plan for the New Year's Eve celebration, which has put an annual strain on the department. Traditionally New Year's Eve and New Year's Day in Richmond meant a full shift for every police officer in the department.

"They pretty much considered the holiday a two-day period and just flooded the streets with officers," said Kelly McCue, formerly a crime analysis supervisor with the Richmond Police Department and now a research scientist with RTI International, who still consults with Richmond. "We fell back a little bit and said, 'Maybe that's not the best way to do things.'"

Last year, Richmond police tried analytics software furnished by SPSS to determine whether the tool could help with the staffing problems caused by the holiday.

"We looked at historical trends and patterns, and Richmond, like many communities, is challenged by random gunfire on New Year's Eve," McCue said. "We decided to go ahead and use a risk-based deployment strategy."

By looking at trends developed from the software, department officials confirmed what they thought they knew: Flooding the area was a waste of resources. Using the software, officials found that the four-hour period between 10 p.m. and 2 a.m. was the critical period and decided to deploy officers in strategic locations at strategic times.

Fifty officers were allowed to have the holiday off, saving the department $15,000. They also reduced citizen complaints by 47 percent and increased the number of firearms removed from circulation by 245 percent.

The results convinced the department that analytics is a key part of its future.

"One thing we realized is that the whole field of behavioral profiling of criminal investigative analysis is based on the concept that crime -- even the most serious, violent crime -- tends to be very homogenous and predictable," McCue said. "If that wasn't the case, we wouldn't have profilers, and we wouldn't have all those cool movies. We'd really be at a loss."

The department now uses analytics early in investigations to get a jump on trying to determine motive, and is beginning to think in terms of models that can predict criminal activity, McCue said.

"Using this type of methodology, we've now been able to characterize certain types of crime, particularly violent crime, using information available at the scene, victim characteristics, things that might have made them a higher risk or lower risk," she said. "That starts to provide insight in terms of the type of person likely to have committed the crime or perhaps motive."

The software clusters information so the investigator can detect behavior patterns and narrow the search, though investigators must rely on their own skills to interpret results, McCue said.

"You're not able to run these algorithms and have it come back and say 'Bob did it."'


The New Tools
Today's analytical tools are getting easier to use, but users must know what data to look for to utilize the tools effectively.

"It's a balancing act," McCue said. "It's important to understand some of the math involved, but the new tools -- especially those that are Windows based -- are highly intuitive."

She said the challenge is acquiring the domain expertise required to develop models that are accurate, valid and useful to the end-user. There are different types of models, and the models can be as complicated or as simple as the user wants.

"If you end up with something so complicated where it's like the [crime occurs on the] second Thursday of each month, then it becomes really difficult to act on it. In some cases, we will compromise on accuracy to have only the variables that are relevant."

Those variables are usually time, location and victim characteristics. In some instances, said McCue, users must be careful not to oversimplify the model. She gave an example of a model designed to predict if armed robberies were likely to escalate into aggravated assault.

The system might come up with a statistic that just 3 percent of armed robberies escalated and could conclude that it was statistically not a factor, which would be dangerous.

"Theoretically you could create a model that's very simple that would say it would never happen," McCue said. She said the hard part is determining what data is available and which model to use.


Going the Wrong Way
Broward County Public Schools in Florida stumbled with its first application of analytics, but is at it again. The district -- the sixth largest in the country with nearly 40,000 employees -- began implementing business analytics about three years ago and has since spent around $25 million.

The district first implemented two SAP business modules where the cost of the software was $3.5 million. The problem arose with implementation of the models, which ended up adding $17 million to the implementation's overall cost. Vijay Sonty, who took over as CIO of the district last year, said the district would have benefited from more planning, but the district is now building some useful analysis models and has learned a few lessons.

"We went the wrong way," he said. "We did a lot of things wrong. We didn't do a needs assessment. We didn't get our users involved. The lessons [we] learned are simply the first things you do before you do a large-scale ERP rollout."

A main sticking point with the district was that it maintained business practices and largely paper-based processes that were 15 to 30 years old, and wanted the software to adapt to those. Specifically, the district had problems matching SAP's payroll system module to processes already in place, and kept spending money trying to make it fit, Sonty said.

"What happened is the customer is always right," he recalled. "[SAP] said, 'Go this way. Adopt SAP methodology.' We said no."

The district spent a lot of money on convoluted systems, and is trying again with SAP -- this time armed with more knowledge.

"With this new implementation, we're doing a detailed needs assessment defining what the current processes are and looking at what processes we can improve."

An important part of implementing analytics is getting buy-in from everyone involved, said SAP's Peck.

"If someone is trying to implement a project and it's just 'that IT project,' that's a recipe for disaster," he cautioned. "In this case, everybody is really on board."

The early benefit is that the district has learned it must streamline and change its practices.

"We're taking a detailed look at the human capital, the financial capital -- do we have the right knowledge and the right management structure to make sure we have the right project leads," Sonty said. "And we're taking a phased approach."

The district is implementing several of SAP's business intelligence modules in phases, including a CRM module, a financial system, an HR system and eventually, a complete ERP portal. This time, the total software cost should be around $6.3 million.

The goals are to determine "business benefits" to each department and manage the data from various departments more efficiently for reporting purposes and performance measurements, Sonty said.

"Previously we didn't have a good inventory system set up. We didn't have a good work order system, and there was a lot of duplication of data and a lot of data entry error. Now we're able to minimize all that because you enter the data once at the source, apply uniform business rules and practices across the district, and combine it into an enterprisewide environment," he said. "The real value for us goes back to the business warehouse. All these modules being integrated, they're able to use SAP's portal environment to slice and dice the data, and run reports."

In addition to the benefits of report writing capabilities, what the software means to the district, in simple terms, is that a user can find pertinent information quickly without having to manipulate multiple systems.

"Previously we had 35 systems to navigate," he said, "and the reports were always wrong."
Jim McKay, Justice and Public Safety Editor Justice and Public Safety Editor