In normal times, fraud is a problem — albeit one that has steadily grown over the past few years. It’s for good reason, too. Fraud is big business, no matter where in the world you operate. The FBI estimates that the cost of insurance fraud (not counting health insurance) is roughly $40 billion a year — costs that are usually passed on to US consumers in the form of $400 to $700 in higher premiums yearly.
However, what happens when we take normal conditions and ramp up volatility and uncertainty? For instance, what happens during a global pandemic, when things become murkier and the volume of insurance claims and applications skyrockets? Risk takes on a whole new meaning, and assessing it becomes more important, but also much more complex. Even worse, the ways risk was assessed even three months ago are no longer relevant when considering the new reality on the ground.
Unfortunately, that’s the reality you currently face — one where your models need to mitigate risk, but where you can’t afford to reinvent the wheel on the fly. The question is, how can you make it better, fast?
The hidden costs of fraud in crises
Risk and fraud are easy to quantify monetarily since there’s a clear number attached to any fraudulent activity. However, these monetary costs only tell one side — albeit an important one — of the story. It’s easy to determine how much a fake insurance claim impacted a healthcare provider. Even so, there’s more to it than just a dollar sign.
For most organizations, detecting fraud and determining risk comes with investments of hundreds of thousands of dollars. Organizations invest significantly in resources, money, and time to create models that can accurately predict and forecast risk. When crisis strikes, as it did in 2020, these costs also expand to include the lost trust and credibility companies face from consumers who were victims of said fraud.
To make matters worse, despite this significant investment, many organizations are still using models that are outdated, clunky, and time-consuming. During the COVID-19 pandemic, for example, when claims are quickly coming in, this can cause delays, errors, and allow fraud to slip through the cracks because models can’t find the right indicators to detect them. Organizations are quickly losing visibility when they rely on historic data that may no longer account for new forms of fraud. It’s time for a new approach.
How machine learning and the right data can reduce risk
In this climate, it may be tempting to swing the other way — to tighten the screws and treat every claim and application as fraud until proven otherwise. Tempting as it is, this comes with its own set of problems. Blocking every transaction, delaying every deposit, and requesting more information from each applicant sounds great, but it means that claims fulfillments will become impossibly delayed, a serious issue in crises.
While many insurance and lending companies still rely on rules-based models to approve claims, this process is somewhat static and can lead to delays. Instead, it’s worth considering a switch to more analytics-based machine learning models that offer dynamic solutions and decision-making. However, it’s not as simple as making the switch — you need to make some important considerations and lay some key groundwork.
The most important foundation you need to establish is having the right data to make faster and smarter decisions. You may have a trove of internal data at your disposal, but this is no longer enough when the situation and conditions are rapidly shifting. Instead, you need to look outside your organization to get better answers.
For instance, let’s say a financial services provider needs to boost its anti-money-laundering (AML) efforts and is building a new anomaly detection model. A standard approach is to look at time-series data and find odd behaviors that could indicate money laundering. While this is a good idea, it produces a high number of false positives when you’re simply running it based on previous positives. Instead, you can add greater contextual data that can look for other, less obvious signs such as company information and financial transaction history.
Keeping with the times
Risk isn’t going anywhere, and crises just make it worse. You need to be able to adapt, update, and reframe your models quickly when disaster strikes to mitigate the threat of fraud, but you also can’t pause to reset your models. The answer, then, is to look for the best ways to improve quickly and without missing a beat. Fortunately, finding the right external data can help you pivot fast enough to keep with the jump in risk.