Aster is an analytic platform with a variety of functions that work together to power discovery analytics. But what does that mean in terms of addressing real business needs? In this post I will attempt to illustrate what we mean when we claim to provide multi-genre analytics at scale, applied to significant business challenges. We have multiple examples that could be used to demonstrate this, but I will focus on just one of the Solutions we have built in the area of fraud detection/prevention.
Every corporation faces a daunting burden to enforce ethical and legal behavior amongst all of its employees lest it be held liable for fraudulent actions. In certain industries, such as consumer finance, the rules have been codified into federal laws aimed at preventing abuses such as usury and redlining. Similarly, the particular seduction of easy profit taking has resulted in regulations for traders and brokers to prohibit insider trading and bribery to coerce deal making.
However, the threat of fraud is not reserved to banking, investing, and other financial services. Organizations of all stripes face a variety of fraud risks. For example, unscrupulous sales tactics to meet incentive targets, or overzealous customer service reps who promise discounts or services that are never delivered.
All of these potential infractions rely upon some degree of fraudulent misrepresentation by an individual on behalf of the organization. The potential damage incurred by the organization can take many forms including: punitive fines, loss of revenue, increased customer churn, erosion of customer satisfaction, brand deterioration, etc.
The approach we have taken using Aster’s Text Analytic functions is to capture the communications logs from and screen them for indicators of potential fraud. It might seem that capturing every email message and transcribing every phone conversations would be quite a challenge, but this is surprisingly easy and inexpensive to accomplish using commodity storage and transcription software.
What jumps to mind next is that screening such large volumes of text data would be a monumental task. However, the scale of the task is actually perfectly suited for machine learning techniques that automate the screening of these countless text files (even in “real time”) against a definable rule set suited for the particular type(s) of fraud the organization is screening for.
Aster’s capabilities provide for intelligent parsing and cleaning of the text files to scrub out the noise and only process the parts that you care about. In the case of correspondences such as email messages, our pre-processing is able to parse and discard echoes, signatures, disclaimers, and other extraneous text, reducing the analytic load by as much as 75%.
Turning briefly to another of Aster’s capabilities, let’s explore how our Graph capabilities allow you to explore the context of each communication. Illuminating the paths that messages traverse gives further insight into identifying message intent. At one extreme, we can key off the distribution breadth and reply rate of a particular message to identify it as unsolicited commercial email (aka spam) that could be removed from further analysis. At the other extreme, a message sent to a single, previously unknown recipient, might warrant additional scrutiny since it would be an outlier relative to expected behavior.
Looking at the results of multiple analytic outputs to better identify potential fraud is an example of how multi genre analytics makes it possible to tackle (and even automate) previously daunting problems. The particular Solution being described makes use of more than the few I mentioned above, but these are the ones that give the biggest synergistic effect when employed in tandem to give you the tools you need to start detecting and preventing fraudulent activities.
If you want to discuss any of the Aster capabilities mentioned in this post or want to hear more about the other Solutions we have built please get in touch.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.