Fraud Detection Engines The elusive Magic Wands
  • Print
  • Share

Advanced Fraud Detection with SignPlus and FraudOne

When a bank thinks that all it needs is a good fraud detection engine. And when a bank should rather be inquiring about image analysis features instead.

Whenever image analysis technology shall be employed to address fraud detection in the context of cheque images, the initial question of banks is: “Is there an engine that can detect fraud on cheque images?” The correct answer should always be: “No there isn’t!” Why is that so? The reason is that this is a subject deeply ingrained into the matters of fraud and the matters of image analysis. But a simple answer would be: Because fraud is not an attribute you can look for. It is a specific kind of acting and more importantly it is the intent. And intent is only visible is it leaves traces.

The most natural idea, to ask for a fraud detection engine, is therefore futile. An intent cannot be detected. But this does not mean the use of image analysis engines is futile. While the fraud intent cannot be detected on an images, the act of perpetrating the fraud is a different story. It is more the subject of asking the right question. Therefore the correct question in the image analysis context would be: “Is there an engine that can help to detect my types of fraud on my cheque images?” And the correct answer will be: “Quite possible! But depends-on!”  It depends on ... quite a number of factors mainly around the question of the fraud itself. What types of fraud is the bank exposed to. What type of traces does such fraudulent activity leave? How can these traces be detected (OCR, ICR, Form Recognition, Pattern Recognition, Data search, and many more)? And how can they be distinguished from traces of regular customer behaviour? And finally, is there an engine that can see those traces and the differences on my cheques? And at what costs? (performance, operational preconditions, false accept rates)

zoom

zoom image: Image engines can find information, detect deviations from a defined norm, and return probabilities on whether certain images elements match predefined values or elements. The detection capability of any engine to find deviations from a given norm does not answer the question whether such deviation is deliberate, or merely the result of an image challenge or simply a customer error. An existing deviation from the norm can be invisible to an engine when the deviation falls in the area of overall "image noise” or when it is smaller than the resolution of the image scan.
Effective and efficient fraud detection is not a mere matter of engines. It is a matter to carefully select the right tools after an intensive analysis of the fraud exposure at hand. And it is beyond that, more importantly a question of how engine results are used and how all kinds of results and additional information are brought together for assessment. The FraudOne Solution Package Family is deliberately focusing on fraud detection as the main goal and purpose of a SignPlus installation. The solution‘s strong fraud detection capability is based on two main features:
  1. The Consilidated Risk Score Engine (CRS) for combining all risk indicator information in a central point.
  2. The GIA common plug-in interface that enables SignPlus to employ any kind of third party engine into the automatic verification contect of SignPlus.
SignPlus is fully equipped to be employed as a cheque fraud detection tool in an image-enabled cheque processing environment. Several fraud-detection oriented installations, especially in the United States show prove for this point. The main factors for success in those fraud detection operations are the availability of multiple data streams plus the results of multiple analysis engines. On top of these risk-indicator-generating mechanisms there is the need for a central evaluation mechanism, which is presented in the SignCheck Consolidated Risk Score engine (CRS). To achieve an efficient as well as effective fraud detection operation all these elements must be considered. For the analysis engines the main focus must be to select something that fits to the current exposure at hand.

SignPlus has a Place for other Engines zoom SignPlus has a Place for other Engines

zoom image: SignPlus has a Place for other Engines

SignPlus has a Place for other Engines

Right at the hearts of the SignPlus application server and verification server structures, among the load blancing features, the remote process control, the error logging and tracing, engien parameter control and results collection, there is the GIA common API. This interface was designed to harbour third-party engines in a way that follows exactly the same administration and control mechanisms as SOFTPRO's own ASV verification server does. Any new engine or new features of an existing engine can therefore quickly be made available to the customer who need this engine's input to their cheque processing or fraud detection environment.