Using Analytics to Derail Fraud Before It Happens

Advanced cognitive solutions could potentially enable organizations to assess fraud and corruption risk on a broader scale, with far greater precision and efficiency, as discussed in a recent Deloitte webcast, The Future of Investigations: Derailing Fraud Before It Happens.

The new approaches to risk management of fraud and corruption are being driven by rapid changes in analytics. “This advanced technology — in the hands of experienced investigators and paired with the abilities of data scientists and other analytics professionals — enables the organization to take the fight against fraud to a new level,” says Don Fancher, a principal at Deloitte Financial Advisory Services LLP and Deloitte Forensic’s global leader.*

There may well be a regulatory benefit as well, since U.S. regulators have invested significantly in analytics to better identify potential fraud schemes, and increasingly expect organizations to utilize advanced technology to monitor potential fraud and corruption risks.

“Regulators will, of course, be interested in an organization’s response to known issues, but also will likely inquire about technology tools and solutions designed to understand the entity’s fraud and corruption exposures based on industry, geography, and business practices. Building preventative and predictive analytics solutions as part of an overall compliance system may serve to reduce potential liability when wrongdoing is discovered,” according to Ed Rial, a principal at Deloitte Financial Advisory Services LLP and Deloitte’s U.S. Investigations leader.*

The Importance of Listening

Identifying vulnerabilities to fraud and corruption through root cause analysis — and taking corrective action by building a monitoring and sensing capability into the organizational infrastructure — can serve to protect the organization by preventing wrongful conduct before it takes hold. “Analytics now present the ability to discern faint signals that may represent outlier conduct across large and diverse data sets,” adds Rial.

“While there is plenty of talk about big data solutions, not everybody understands what that means: It requires the organizational openness to hearing what the data is saying, rather than heeding preconceived notions about key risk indicators,” notes Doug Veivia, vice president at Prudential Financial and a member of Prudential’s international insurance compliance team. “To be ready to accept what the data is indicating is critical. Sometimes the toughest challenge is having the right people with the requisite business knowledge and openness to approaching problems and being informed in different ways. And data analytics helps address that challenge.”

Building an Analytics Program for Detection of Fraud and Corruption

When starting to build an analytics program for fraud detection, it’s critical that the organization know what it is trying to solve for, settle on key questions to address, and determine the scope: If the analytics program seeks to solve every issue the organization faces, it could start running in circles with no output.

“Building an analytics program requires strategic planning, a road map and a combination of people, technology and well-thought-out goals,” says Satish Lalchand, a principal at Deloitte Transactions and Business Analytics LLP and Deloitte Forensic’s analytics leader.** “There is no single tool that is a silver bullet with respect to the issue. Rather, is it is a solution built around a strategy,” he adds.

Four basic principles govern the building of a new analytics program (see chart below). But it is important to keep in mind that such programs are not just about data scientists, technology, software, or infrastructure; but rather the people who have the insight. “A lot of this work still involves intuition and the experience of having worked in the industry for long time and gaining the core forensic skills required to understand when a problem is rising,” says Rial.

“To get started, an organization might consider performing a self-assessment in order to establish a baseline of where it stands with respect to the type of analytics it’s using, if it’s using any at all,” suggests Lalchand.

In general, there are six broad analytics techniques that organizations can consider when building an analytics program: Rules, anomaly detection, machine learning, visual analytics and dashboards, text analytics, and network analysis.
 

An organization can use analytics techniques and build models in many ways, and using a combination rather than just one may be better. “For example, there are many rules-based approaches to flagging certain behaviors that are designed to address a specific compliance or regulatory risk within the financial services industry. But for the most part, these approaches are inflexible; they flag only behaviors that are relatively easy to circumvent, and they are limited by the human bias that is deciding what the rule is,” notes Fancher.

As an example of the benefits of using more than one analytics technique, one organization wanted a model that not only would identify customer complaints, but also predict when one might occur. The model began with identifying customers who had submitted complaints, and via an application of predictive analytics and text analytics, the model was able to analyze various aspects of the behavior of those employees who were the subject of the complaints. The model also looked at data such as commissions, earnings reprimands, and compliance issues, and, ultimately, could predict whether an employee might be associated with a certain type of fraud complaint.

“From a financial services perspective, two main goals tend to drive an organization’s selection of which types of analytics capabilities to use,” notes Veivia. “The first is protecting its customers, and the second is driving sales growth. Having greater sensing ability can allow the organization to identify troubling behaviors earlier — before there is real customer impact and then regulatory — to inform how risk-monitoring rules should change going forward,” he adds, noting that financial misconduct or criminal activity tends to evolve over time.

“As we place procedures and controls in place, the conduct will evolve to circumvent those controls. So what’s best for risk management is having an ability to detect changing behavior that can then be identified or sensed earlier, and addressed,” says Veivia.

Arriving at mature analytics is a measured journey. “Once an organization understands the end goal, it can build a strategy and a roadmap for the different capabilities it seeks,” suggests Lalchand. “The organization could then take concrete steps, such as conducting pilots, working with smaller chunks of data, or experimenting with models before making any larger decisions around tools and approach,” he adds.

It’s important to note that when an organization does obtain findings through analytics techniques, these findings merely represent leads. Further confirmation is needed to identify that fraud is actually occurring. “But once fraud is confirmed, the organization can feed the cases it has researched back into the model — and thereby make the model more agile and smarter,” says Rial.

*Ed Rial and Don Fancher are Deloitte Risk and Financial Advisory principals in the Forensic practice of Deloitte Financial Advisory Services LLP.

**Satish Lalchand is a Deloitte Risk and Financial Advisory principal in the Analytics practice of Deloitte Transactions and Business Analytics LLP.


This article was originally published on wsj.com