As April 15 looms and tax season starts entering its frenzied home stretch, individuals and corporate accountants contemplating their tax returns may find themselves wondering just how much they can, let us say, fudge the numbers.
Don't be tempted. In recent years, the Internal Revenue Service has become markedly better at spotting potential tax fraud and more aggressive in pursuing it.
In 2006, the agency collected a record $59.2 billion via 1.4 million audits. Revenue has grown during the past seven years and is up 75% since 2000.
In particular, the IRS has increased its targeting of tax cheats in the middle-class (between $25,000 and $100,000 in annual income) and super-rich (greater than $1 million a year) brackets. It audited 436,000 middle-class households and individuals in 2006, three times as many as in 2000. The chances of being audited rose commensurately, to 1 in 140 in 2006 from 1 in 377 six years earlier, according to The New York Times.
Meanwhile, the above-$1 million club faced a 1-in-11 chance of an audit in 2006, up from 1 in 20 three years earlier (download PDF).
At the same time, the IRS has been able to steadily reduce the number of audits of honest taxpayers, according to Jeff Butler, director of research databases at the IRS.
Butler oversees the agency's research data warehouse, key to the agency's recent improvements.
"What we do is, in some respects, cutting-edge," said Butler, with the restrained pride of a longtime bureaucrat (he has been at the IRS for a total of 15 years, with an additional five-year stint at the U.S. Department of Transportation). "We're getting to be as sophisticated as the largest credit card companies or banks."
A world-class data setup
Butler may be too modest. Unifying all tax returns and related information from the past 10 years, the 150TB Compliance Data Warehouse (CDW) is comparable in size to the largest known databases in the world, such as those run by YouTube, AT&T and the CIA.
It's not just the CDW's size that impresses, but also its capabilities.
IRS researchers can use it to "search and analyze hundreds of millions or even billions of records at a time, so we can answer questions, look at trends, do simulations and optimization modeling," Butler said. And those analyses, rather than taking weeks or months as they would have in the past, now take hours or days, he said.
Using the CDW, the IRS was able to discover areas where tax cheating had become rampant, such as the Earned Income Tax Credit or small-business tax shelters, Butler said.
Researchers have also used the CDW to discover or confirm who is at risk for falling behind on their tax payments. For instance, young graduates laden with college debt are particularly susceptible.
Naturally, the system has its limits. It doesn't assist IRS agents in the process of actually conducting audits. And it isn't the system that returns information to individuals or corporations seeking answers about particular returns.
"We aren't open to the public. And we don't manage any transactional data stores. The way we deliver value is behind the scenes," he said.
Putting it together
Like any large federal agency or corporation, the IRS stores data from many sources -- including legacy mainframe databases, Oracle databases and flat files. Linking them to do any sort of business intelligence work a decade ago was a "nightmare," Butler said.
For the CDW, Butler's research group chose 10 years ago to adopt Sybase Inc.'s then-new IQ Analytics Server. Unlike most horizontally minded databases, IQ stores data in column-based tables, which can be slow for writing data but quick at reading it -- making it perfect for applications such as data warehousing.
"We were met with a lot of skepticism," Butler recalled. "'Why don't you go with IBM or Informix?' people asked me. It looked a little risky, too, since Sybase IQ wasn't part of the approved enterprise architecture. Eventually, I was told that we could use IQ for research, but we were going to keep on mostly using Oracle and IBM DB2."
Actually, the research division also uses Microsoft Corp.'s SQL Server to store all of the metadata for the data warehouse and the rest of the agency. Managing and cleaning all of that metadata -- 10,000 labels for 150 databases -- is a huge task in itself, Butler said.
Better hardware, improvements to IQ and, most of all, faster and cheaper storage have combined to boost the CDW's performance.
"When we started, it took us six to eight weeks to load one year's worth [15 to 20TB] of tax returns. It takes four hours today," Butler said.
Today, only about 500 researchers -- mostly from the IRS, some from the U.S. Department of the Treasury -- are allowed direct access to the CDW, using Hyperion Solutions Corp. business-intelligence query tools.
But Butler's team is starting to build analysis services that would calculate and publish summary and trend statistics to wikis, blogs and SharePoint sites inside the IRS. That would make it easier for the IRS to fulfill data requests from federal agencies.
The ultimate goal is to make the information available to the public, similar to how the U.S. Census Bureau publishes copious demographic data to its Web site. That, Butler says, would be "the pot of gold at the end of the rainbow."