Machine Learning – One New Weapon To Combat Fraud

Dan Wellers

Trillions of dollars are squandered every year as a result of fraud – around $3.8 trillion, in fact. The typical organization loses five percent of revenue annually to fraud, and eight out of 10 executives said their company had fallen victim to at least one instance of fraud. Every type of fraud has risen in recent years and is in the double digits today – from financial fraud and market collusion to vendor and supplier fraud to theft of physical and information assets.

And that’s just the fraud that companies know about. In some areas, like the supply chain, malfeasance often goes undetected. Nearly half (47%) of executives and managers did not even know whether their company had experienced fraud, waste, or abuse in its supply chain, according to a recent survey conducted by Deloitte.

In the digital age, where corporate transactions and interactions can take place in milliseconds, identifying or preventing fraud is simply beyond the realm of human capabilities. The traditional technology solution to fraud detection is to apply rules-based analytics – 90% of online fraud platforms (many of them at work in the financial services industry) – still use this method. Such systems flag credit card purchases that take place outside a customer’s country of residence, for example, or unusual payments to offshore suppliers. The problem is that those rules are developed by humans, based on data and trends but also intuition. While it’s somewhat effective, the approach is costly, slow, leads to false positives, and fails to keep pace with emerging trends. It’s no surprise then that 74% of executives say they need to improve their current anti-fraud procedures, according to an EY survey.

Being able to detect bad actors more efficiently and effectively – every minute an instance of fraud goes undetected, a company’s losses increase – is a top priority. Enter machine learning, which uses mathematical algorithms to analyze bigger, more complex data sets and deliver faster and more accurate results. Rather than robotically isolating certain types of transactions, fraud detection solutions enabled by machine learning analyze historical transaction data and learn from them to build models that can detect fraudulent patterns even as they happen in real-time. Using transaction data, these self-teaching systems can model a vendor’s typical transaction profile (dates, quantities, types, and values of goods). When a new purchase order is submitted, the system generates a fraud score along with a flag if the score exceeds a certain threshold.

Machine learning capabilities are particularly suited to this risk management challenge. They can:

  1. Detect fraud in real time
  1. Learn from trends and identify emerging fraud patterns quickly
  1. Integrate and analyze changing, unstructured, and fast-moving data in ways that humans alone cannot
  1. Run multiple algorithms in parallel to increase fraud detection capabilities
  1. Identify rare or never-before-experienced fraud events
  1. Automate tedious tasks and free fraud examiners to focus on the forensics work that requires human analysis
  1. Predict future behavior and fraud patterns

While machines can better perform the arduous task of dispassionately sifting through massive sets of structured and unstructured data for fraud patterns, it’s important to remember the critical role that humans play and how company culture must support it. Fraud is quite often exposed when an individual is empowered to speak up. For example, while 55% off companies have some kind of whistleblower policy in place, those processes are not always effective, according to an EY report. In the UK, where regulators began adopting new tools to support and encourage individuals to come forward, such efforts appear to have paid off. Last year the number of fraud incidences reported by whistleblowers was larger than those self-reported by companies.

To better identify, prevent, and limit the impact of fraud in the digital age, companies must invest both in new artificial intelligence capabilities and support for identification of troubling trends by individual employees. By combining the best of man and machine, enterprises can be better equipped to keep pace with this advancing and costly threat.

For more on the changing role of people and organizations as artificial intelligence takes hold in the enterprise, see An AI Shares My Office.

This blog is the first in a six-part series on machine learning. 

About Dan Wellers

Dan Wellers is the Digital Futures Global Lead and Senior Analyst at SAP Insights.