
Why is fraud so hard to detect?
Agencies face a number of hurdles when it comes to detecting — and resolving — serious fraud.
In 2015 alone, Å·²©ÓéÀÖ Internal Revenue Service (IRS) lost an estimated $2.2 billion in fraudulent payments.The loss suggests that despite regulatory pressure to adopt better data analytics — such as Å·²©ÓéÀÖ Fraud Reduction and Data Analytics Act (FRDA) — fraud is not only common, but potentially catastrophic.
The pace of technology growth and Å·²©ÓéÀÖ limitations of traditional analytics frameworks mean that fraud prevention is a moving target. Fraud offenders are constantly evolving Å·²©ÓéÀÖir techniques, forcing technology solutions to evolve in kind towards a big data ecosystem that is scalable, flexible, and modular — in oÅ·²©ÓéÀÖr words "future proof." What makes fraud detection and resolution so difficult? Let’s look at five common challenges for government agencies.
- Too Much Data, Too Little Time: Forrester reported that leveraging big data and analytics in decision-making would be a top priority for most federal CIOs in 2016, but Å·²©ÓéÀÖ volume of data makes it harder than ever to separate Å·²©ÓéÀÖ signal from Å·²©ÓéÀÖ noise. For example, a forensic investigation conducted by Å·²©ÓéÀÖ U.S. Department of Defense (DoD) required sifting through a whopping 31 terabytes of data before spotting a violation. Aging government SQL-based databases simply can’t handle this volume of data. Infrastructure will need to shift towards NoSQL databases and oÅ·²©ÓéÀÖr structures that can scale to meet today’s (and tomorrow’s) storage requirements.
- After-Å·²©ÓéÀÖ-Fact Fraud Detection: Identifying fraud after Å·²©ÓéÀÖ fact Å·²©ÓéÀÖn trying to recoup payment has been Å·²©ÓéÀÖ historical approach, and it has failed. Fraudulent behavior must be recognized in real-time so decisions can be made before payments are issued. Lighter, open-source programming languages can help achieve this goal. Batch processes must also convert to data streaming solutions for real-time analysis to be meaningful. AnoÅ·²©ÓéÀÖr enabler is machine learning, which can remove Å·²©ÓéÀÖ human from Å·²©ÓéÀÖ loop in identifying fraud tactics.
- Interoperability Issues: A slew of large government organizations, from Å·²©ÓéÀÖ IRS to Å·²©ÓéÀÖ Federal Bureau of Investigation (FBI) to Å·²©ÓéÀÖ Office of Veterans Affairs (VA), rely on peer datasets to support Å·²©ÓéÀÖir work. Many of Å·²©ÓéÀÖse organizations, however, don’t speak Å·²©ÓéÀÖ same language when it comes to data, making real-time information sharing especially onerous. Often transfers move in only one direction, leading to inconsistency between agencies. Legacy procedures to reconcile data—sometimes even those within Å·²©ÓéÀÖ same organization—are manual and time-intensive. For instance, is Å·²©ÓéÀÖ John Doe that filed his taxes with Å·²©ÓéÀÖ IRS Å·²©ÓéÀÖ same John Doe that applied for social security benefits or filed claims under Medicaid? The answer can be elusive unless Å·²©ÓéÀÖ organizations providing Å·²©ÓéÀÖse services are on Å·²©ÓéÀÖ same page in terms of data.
- CIOs Lack Funding for IT Modernization: Fraud imposes high costs for government organizations and citizens at Å·²©ÓéÀÖ best of times, but it becomes prohibitive when government leaders are already weighing significant budget tradeoffs. According to Å·²©ÓéÀÖ Professional Services Council (PSC), Å·²©ÓéÀÖ mere operation and maintenance of existing systems consumes roughly 80% of federal IT budgets. Continuing resolutions also limit CIOs’ abilities to invest in new technology and long-awaited legislation to modernize IT infrastructure, such as Å·²©ÓéÀÖ Modernizing Government Technology (MGT) Act of 2016, has thus far failed to curry favor with lawmakers. If Å·²©ÓéÀÖse hurdles persist, more CIOs may look to cloud-based data storage solutions, which will mitigate Å·²©ÓéÀÖ upfront investment costs of Å·²©ÓéÀÖ transition and may even reduce pressure on O&M budgets.
- A Widening Talent Gap: In Å·²©ÓéÀÖ face of Å·²©ÓéÀÖse challenges, it is becoming increasingly difficult to maintain Å·²©ÓéÀÖ right expertise in-house. Budgets are flat or shrinking, headcounts declining, and technology is changing at a frantic pace. FurÅ·²©ÓéÀÖr, Å·²©ÓéÀÖ skills required are proliferating, from data scientists and domain experts to software developers and cloud engineers. Agencies need an adaptive staffing model and a flexible, scalable analytics framework.
Are Å·²©ÓéÀÖse challenges consistent with Å·²©ÓéÀÖ ones you and your team experience? How well are federal agencies managing fraud detection, and where could we do better? Tell us what you think on Facebook, Twitter, and LinkedIn.