Tax evaders beware: NY state has an app for that
Analytical engine, developed with IBM, helps New York recover millions in unpaid taxes
- By William Jackson
- May 11, 2010
The New York State Department of Taxation and Finance collects about $3 billion a year in taxes, and because it's a major revenue center, its budget has not been shorted during the recent recession.
“We’ve been pretty lucky so far,” said William Comiskey, deputy commissioner of tax enforcement. “We’ve grown over the last two years by about 20 percent overall, and we hope to maintain that growth.”
However, ensuring that every tax dollar owed is being collected becomes even more critical to the state during this economic downturn. The department processes about 160,000 returns a day during its busy season, and each has to be reviewed for accuracy and occasionally followed up with audits and collection activity.
“Even with the additional resources, we are still under pressure,” Comiskey said.
NY turns $10 million investment into $1 billion in savings
That pressure is being eased somewhat through a collaboration with IBM Research and IBM Global Business Services that has produced automated tools to help audit returns and collect delinquent taxes. The effort has paid off for both the state and IBM. Comiskey said his office has saved $1 billion since 2003 by improving its audit process with the Tax Audit and Compliance System developed with the company, and the office is on track to bring in an additional $100 million in delinquent taxes during the next three years using the Tax Collections Optimizer (TCO) that started operating in December 2009.
“Everything we’ve seen so far leaves us confident that those projections are sound,” Comiskey said.
For its part, IBM gets to announce TCO as a new offering in its Global Business Services’ fraud and abuse management portfolio.
“The Department of Taxation and Finance has had a long relationship with IBM,” dating to a 1999 modernization project in the department, said Chid Apte, director of analytics research at IBM’s Thomas J. Watson Research Center.
Matching the department’s needs to available IBM technology produced commercial tools such as the Tax Audit and Compliance System, which Comiskey said is extremely successful. TACS does not replace human auditors, but it does an initial examination of returns and flags those that need to be followed up by an auditor. Automating the process has saved money through improved efficiency, and it helped increase collections by improving the ability to spot underpayments or erroneous refunds. With this information, the next step was to improve the collection process.
“About three years ago, there was a desire to improve post-audit activity,” Apte said. “There was a fairly large bucket of money the state needs to recover year after year.”
Collections typically are handled by county-level offices in a static but complicated system. “It was very canned,” Apte said. “No matter what kind of recovery they were going after, it was the same process.”
IBM began applying data analytics tools to find a way to optimize the collection process. In analyzing historical collections data, “we found that different sequences of action resulted in optimized recovery for different types of parties,” Apte said.
This was not a surprising finding. The challenge to using this information was to develop specific plans for each case that would minimize the expenses and maximize the returns. IBM already had developed the analytical engine that could examine historical data and figure success and failure rates under sets of sequential actions. The analytical engine has been used in industries such as marketing, financial institutions and transportation. But in some ways, tax collection was more complex. There are many types of individual and corporate taxpayers, with different behaviors, and any decision engine had to understand the organizational, legal and budget constraints of the tax collection environment.
“Analytics can be used in the lab, but to make it practical in the field is a nontrivial activity that took a lot of hands-on work by the research team to make it happen,” Apte said. “Over the last two or three years, we had two to four researchers working with the New York people to understand the collection environment.”
TCO develops a collection plan for each case, after analyzing individual taxpayer data, such as the amount owed and past payment history. It creates a workflow that looks for areas that promise the best use of resources for the agent to focus on. It looks at data for the whole state and determines whether each case should be handled out of a central office or a local office and whether the next contact should be by letter, a phone call or a personal visit, weighing the costs of each action against the probabilities of success and the amount at stake.
After a period of testing, TCO went into production in December 2009, although with limited functionality. “We think that when it is fully implemented, it will increase our collections by about $35 million a year,” Comiskey said.
Because the system’s decisions are based on past experience, the process and success rate should improve over time. “This software can learn,” he said. “That takes my breath away.”
Some cases now are being handled in the field that would not have been before, because it was determined that a personal visit would be more effective than a letter or phone call. Other cases do not require a visit and can be handled more efficiently from a central office.
Combining automated audit and compliance evaluation with an enforcement decision and workflow system has produced what Comiskey called a “spectacular system that moves enforcement into the processing stage. It’s a change that is leading to greater efficiency.”
William Jackson is a senior writer of GCN and the author of the CyberEye blog.