Reference no: EM133545410

Case Study: Cracking Fraud with Government's Big Data and AI

Driving around Massachusetts, you might notice an RV frequently parked in your neighborhood with Montana license plates. Perhaps it is someone visiting from out West, but it might also be a Massachusetts resident who bought the RV online and registered it in Montana, which has no sales tax. The RV owner saves thousands of dollars, but Massachusetts loses the tax revenue. This kind of fraud is illegal but difficult to catch. A Massachusetts state agency, for example, would not have access to Montana's vehicle registrations and would be unable to match them up against Massachusetts' tax forms or employment records. The data is there but not integrated into big data that can be analyzed.

The healthcare system is especially plagued by fraud at both federal and state levels. Analysts estimate that fraud and abuse cost billions annually, but problems often go unnoticed because they are difficult to identify. Only a small percentage of fraudulent cases are discovered, and the detection often happens so late that the funds cannot be recovered.

Analytic techniques are tools that can spot suspicious activity and unusual patterns. The potential to reduce government waste and fraud, in general, is enormous. Combined with big data, these tools can arm investigators with ways to track fraudulent billing patterns buried in millions of legitimate claims, picking out unusual trends that no human working alone could see.

For example, Health Care Service Corp. (HCSC) implemented a fraud detection system that paid off almost immediately. An allergist in Illinois was submitting fraudulent bills, but the individual amounts were never high enough to trigger any suspicion. Something was amiss, however, and the analysts for the insurance company were able to compare what other allergists were charging for the same procedures. The results helped uncover an $800,000 scam.

With access to big data from multiple sources, fraud detection systems can spot a large variety of suspicious activities that need investigation, particularly if data can be drawn from state and federal databases. In health care, for example, such systems start with rules that flag unusual behavior in near real-time, such as when a provider bills for many services in a short time window or when a person enrolls in Medicaid in more than one state.

With artificial intelligence and machine learning, the software can also learn from the data to build more sophisticated models, especially as the tools have access to more and more data. They can examine relationships and rapidly identify suspicious cases with improving accuracy so investigators can follow up immediately. Analyzing these patterns can reap huge benefits and uncover large criminal gangs even if each individual transaction on its own does not trigger any flags.

The near-real-time analysis made possible by AI is especially important because of the need to spot fraud before any claim is paid. It is much easier to deny or delay payment than to recover funds already paid out. However, the time window is short, given the pressure on payers to reimburse quickly. With these analytical tools, fraud detection systems can operate quickly enough to catch fishy claims before they are paid.

The challenge of reducing fraud is daunting for the 50 states, which have different information systems, and all the counties that maintain their own records. The goal is to make a big data view of citizens, one that would, for instance, inform a state agency if someone receiving benefits purchases a luxury vehicle in a state with no sales tax. With states and counties struggling with budget woes, the drive to catch fraud is strong.

Will "big data" become "Big Brother"? Privacy advocates voice concerns over the growing access to big data across government agencies, particularly as analytics becomes so sophisticated and ways are found to integrate the data to paint a meaningful and comprehensive picture of each person's financial transactions and government benefits. Disclosures about the extent of the government's electronic data gathering for national security have intensified those concerns. Balancing privacy and ethical concerns against the need to reduce fraud will be particularly important.

Discussion Questions

• How could data mining and analytics be used to detect fraud in health insurance claims?
• How could private insurance companies and public government agencies collaborate to combat insurance fraud?
• What types of business skills would be necessary to define the rules for and analyze the results from analytics?
• What business processes are necessary to complement the IS component of artificial intelligence?
• Do you think "big data" will become "Big Brother"? What concerns (if any) do you have about using artificial intelligence in government?