What to expect when implementing NetGuardians
With pre-configured adaptors and software, NetGuardians’ fraud-mitigation software could not be easier to implement, allowing banks to prevent more fraud within weeks of signing up, writes Michaël Gingins
It takes just a few weeks to install NetGuardians’ fraud mitigation software, upgrading a bank’s protection to best in class not just for now but well into the future.
With more than 50 completed implementations at banks big and small, we have the process down to a fine art. Our relationship starts with the first approach by a bank and, once we’ve gone live, it never ends – 100 percent of our customers come back again and again. And the longer we work together, the deeper and better the relationship becomes, helping prevent more fraud while ensuring the best possible customer experience.
Questions from potential clients usually focus on our software’s capabilities in terms of analytics, algorithms, reporting, customization and the technology behind it – Big Data processing, distributed computation, profiling and machine learning, as well as real-time reporting. Banks also want to know what they can achieve in terms of fraud prevention – the answer is internal and external, lone wolf and collusion, professional criminal gangs and individuals – and what they will need to do to prepare for our solution.
When it comes to the last point, the answer is very little. Banks can continue their standard operations while we implement our software. Disruption really is minimal, particularly for institutions already running a digital core platform. We have preset connectors for core banking systems such as Finnova, Avaloq, and Temenos. Further, we have open APIs, which make interfacing very simple. However, a bank will need to ensure it has commodity hardware capacity to allow it to process in real time the very high volumes of transactions it experiences.
All implementations start with a validation phase, where we run workshops with the business users to understand their organization, their existing pain points and what they are trying to achieve. We look at what the bank is already doing to mitigate fraud – usually rules-based processes such as validation for transactions above specific ceilings, two-party validation, passwords etc. From this we develop a functional specification design, which sets out top to bottom what our machine learning software will do.
It’s at this stage that we work on controls and workflows, and look at what the bank will need to do to achieve the desired results. The focus here is on ensuring our software is fed the right data to create 360-degree customer and employee profiles against which all transactions are compared to find anomalies and flag only truly suspicious behavior.
For smaller banks, this process is usually straightforward and our pre-packaged solutions often prove useful. Larger banks, with multiple databases running legacy systems, are more likely to require tailored implementations. Those running digital core platforms from Temenos, Finnova and Avaloq will have easily accessible data and it’s a simple question of using the right adaptor or our pre-configured system.
As with any software implementation project, the closer to out-of-the box it is, the faster it is. On average, our packaged solution can be deployed very quickly – taking between one and two months from start to finish, including loading the bank data and two rounds of calibration and review.
Calibrating and refining
Typically, we use six to 12 months of historic bank data to set up the software. During this process we calibrate the weightings assigned to each data stream so that the maximum number of fraud attempts can be spotted without throwing up a plethora of false alerts.
By using historic data, we can compare frauds identified by our software against those the bank is already aware of. In this way the software learns about the bank, its staff and customers and builds the profiles. The more data it crunches, the more it learns and the better it performs in terms of highlighting only anomalous behavior.
Once the first round of calibration is complete, we run the software again for a number of weeks before reviewing the results and recalibrating as required ahead of the go-live.
Live, our machine learning software is used to monitor all transactions in real time. Banks are often surprised at how quickly the software starts to spot new frauds – perhaps between staff or a relationship manager defrauding a customer – or new fraud types resulting from emerging fraud scenarios.
This phase runs for a few weeks and monitors the alerts, false positives and the types of fraud detected. This information is used to refine the machine learning solution further.
Experience shows that banks running NetGuardians’ software quickly become more efficient at spotting fraud, with 83 percent fewer false alerts, and spend 93 percent less time investigating fraud. But that’s not all. In one trial, we found that our software identified 18 percent more fraud than the bank’s previous rules-based fraud-mitigation processes. All this without having to train staff to read specialist reports and dashboards. Indeed, automated reports come in a human readable language so they can understand quickly why the machine learning algorithms have raised an alert.
Fraud is not a static problem – because the fraudsters are constantly trying new methods. In response, we regularly update our library and recalibrate clients’ controls annually.
Feedback from banks is extremely positive. They find the implementation process unobtrusive, the software easy to use and that it spots and stops more fraud.
Furthermore, their customers appreciate not being bothered as often to verify bona fide transactions, and that when they are contacted it is for a transaction that is really out of character. Implementing NetGuardians’ fraud-mitigation software is a win-win for everyone – apart from the fraudsters.