Unlocking Business Value From Improving SAS Fraud Solutions
My colleague @saurabhbatra recently wrote on the development of a fraud model with a customer. He added that the customer, a large bank, was already utilising a fraud management solution but had noted a significant rise in the number of prospective frauds. As a result, the bank requested more assistance in order to increase the accuracy of its fraud detection. As an addition to SAS Fraud Management, a new machine learning model was developed.
The additional model resulted in a 100-fold increase in value for the customer. Saurabh's post focused on analytics team learning. What can we learn, however, about how clients might get commercial value by enhancing fraud solutions?
Six key areas
There are six broad areas that were affected by the new fraud detection model, and which can help to deliver business value. These were:
- Increased fraud detection accuracy
The new model was significantly more accurate. In other words, it enabled the bank to identify more genuine fraud cases. It also reduced the number of false positives, or the number of times that the model highlights genuine transactions as potentially fraudulent. Taken together, these mean that genuine frauds are more likely to be identified, with less time wasted on investigating genuine transactions or customers. This therefore leads to better decision-making and improved fraud prevention.
- More efficient fraud detection
The improved accuracy of detection, and particularly the reduction in false positives, improves efficiency of investigation and fraud prevention. In particular, it reduces the number of suspicious cases that need investigating. This enables investigators to focus more accurately on the relevant cases. This reduces the overall time needed for investigations, and maximises the benefits of investigation.
- Higher productivity
In turn, improvements in efficiency enable teams to target cases for investigation more effectively. This translates to better use of resources and more effective fraud prevention efforts—and to a more productive use of staff time. Investigators are not ‘running to keep up’, but can focus on the high-risk cases with more confidence.
- Faster risk-based decision-making
The model described by Saurabh provided for real-time scoring of all transactions. This facilitated faster risk-based decisions. The bank could respond promptly to potential fraud cases, investigating pre-emptively rather than after the fact. This minimises the potential impact of fraudulent activity.
- Improved customer satisfaction
The improved fraud detection and reduced number of false positives meant that legitimate customer transactions were less likely to be incorrectly flagged as fraudulent. This, in turn, meant that genuine transactions were more likely to proceed quickly, improving the customer experience. Customers also benefit from fewer losses to fraud, because banks do not need to recoup the losses elsewhere.
- Adding value to an existing solution
The machine learning model described by Saurabh was designed as an add-on to the bank’s existing fraud management solution, SAS Fraud Management. This is a good way to add value, because it means that existing solutions do not need to be replaced, with the consequent staff training and potential teething problems. Instead, they can be enhanced with targeted, specialised models to address customer-specific problems. This provides a significant value-add for the customer.
Indicating value
The bank measured the value of the additional model by looking at six key performance indicators (KPIs).
- The first was increased fraud detection rate, which effectively looks at the bank’s ability to identify and detect fraudulent transactions accurately.
- The second was a reduced level of false positives, or legitimate transactions mistakenly identified as fraud.
- The third was focused investigations: the ability of investigators to focus on more relevant cases, improving their efficiency and effectiveness.
- The fourth KPI was reduced time to decisions about potential frauds, and how the model facilitates faster risk-based decision-making.
- The fifth was team productivity, and how the model enables the team to handle more cases effectively.
- Finally, the sixth KPI was reduced customer complaints, reflecting the model’s ability to reduce the false positive rate, and also increase detection of genuine frauds. All of this speeds up the transaction time, and therefore creates a better customer experience.
Together, these six areas have a significant impact on the customer’s bottom line. There are three main improvements on the ‘input’ side. The first two, reducing the volume of fraud and the number of false positives, are both process improvements. The increase in productivity is a ‘people’ improvement. They contribute to increased revenue, better customer satisfaction, and an increase in brand loyalty—all of which add up to greater profits.
- Sidharth Khona (Business Advisory & Business Value Lead India)
