Human Expert Knowledge or Predictive Algorithms - why not use both at the same time for fraud detection at Public Sector organizations

[Former Member]

Many Public Sector customers are looking to improve their control systems to reduce fraud and improve compliance. One of the challenges they face is to figure out how to build business rules which will create only alerts for fraudulent/non-compliant transactions but do not create any alerts for flawless transactions.

While most customers already have some sort of control system in place which is based on rules which reflect the professional experience of employees or external consultants, few have dared to leave the determination whether an alert shall be raised for a suspicious transaction to a mathematical algorithm.

Why was that?

Predictive Algorithms – although an art which has been successfully applied to credit cards fraud detection for more than 20 years - few public sector agencies have had the human skills nor the technical capabilities to apply such means.

What is changing?

Many customers are modernizing their old legacy systems and may see advantages in deploying transactional systems with in-built predictive algorithm capabilities and thereby avoiding to have extra systems for analytical processing.

Furthermore, there are modelling tools on market which allow to apply predictive algorithms at data sets which do not require a data scientist degree (for example SAP Predictive Analysis – automated analysis)

However, both these trends do not eliminate the human factor that predictive algorithms which are normally not understandable by a non-data scientist trained mind  and thereby might not be accepted by the employees which are providing the human expert rule knowledge. Often people may also consider this new “things” a threat to their expert status. 

While predictive analysis has been historically proven to be an extremely useful tool automating  fraud detecting , there is no guarantee that predictive algorithms are always superior to human expert rules. Therefore this author argues that ideally an introduction of predictive analysis should be accompanied with a continued application of human expert rules. This will allow the organization to show the benefits of both, plus it allows for time to slowly show hesitant staff that “predictive rules” might show value and that the application of latest technology should rather be embraced than dismissed.

Albrecht Weiss

Solution Management Public Sector

SAP SE