On Tuesday, April 17, 2007, in Palo Alto, SDForum’s Business Intelligence SIG hosted Detection of Irregularities in Accounting Data. It was presented by Stephen Bay from the Center for Advanced Research at PricewaterhouseCoopers LLP.
Stephen Bay is the analytics lead on Project Sherlock. The goal is to detect risks of potential financial statement error or fraud by uncovering anomalous activity in a company’s general ledger.
Accounting frauds like Enron, WorldCom, Adelphia or HealthSouth have destroyed trust in financial statements. A company’s financial results intentionally misrepresented by billions of dollars can disrupt markets and whole economies. The actions of few can cost thousands of innocent people to lose their jobs or lifesavings.
Responding with Sarbanes Oxley and other laws, regulatory bodies mandate auditors perform analytics on detailed financial data to discover misstatements. For a large auditing firm like PWC, this may mean analyzing millions of records from thousands of clients. Fraud detection normally uses logistic regression models to look at financial statement ratios. An expert named Cecchini can even perform textual analysis based on word frequency. Sherlock takes a new direction.
Stephen Bay discussed techniques for automatic analysis of company general ledgers on such a large scale to identify irregularities. These might indicate fraud or just honest errors, but still need additional review by auditors. These techniques are now in a prototype system called Sherlock combining aspects of classification, anomaly detection, and signature identification.
Sherlock looks at a companyâ€™s the general ledger and not their financial statements, which at best are aggregated data simplified for investors. This gives more data to analyze, compare and contrast when looking for patterns. Any change in pattern, new or unusual sales, purchases or expenses can be further examined.
Sherlock is powerful, but it just another auditing tool. It can augment, not replace traditional auditing techniques. Bay and his team are trying to increase Sherlockâ€™s dependability to reduce the chance of a false positive or negative result but they cannot statistically eliminate it. In the end, no matter how the data is processed or displayed, a human will have to look at it and make a judgment call.
PWC Researchers David Steier and Markus Anderle were also available to answer questions about Sherlock.
Copyright 2007 DJ Cline All rights reserved.