racetrack.ai

Blog

Admin / April 10, 2019

Fraud Detection & Security: The AI Value Proposition

Fraud Detection & Security: The AI Value Proposition

Going by the usual definition in layman terms, ‘Fraud’ includes a range of illicit practices involving intentional deception or misrepresentation. As per the Institute of Internal Auditors’ International Practices Framework, Fraud is any illegal act characterized by deceit, concealment or violation of trust.

Having set the definition, let us now touch the industry this blog would specifically discuss – the BFSI sector, a sector where money is involved  with financial management being the core. Any fraud in this sector not just costs the firm its revenue and efficiencies, but also a loss of goodwill which mars the future growth prospects as well.

As shocking as it may sound, but financial organisations across the world lose about 5% of their annual revenue to fraud. McAfee recently, in a report estimated that cybercrime costs the global economy a whopping $600 billion per year or as can be put alternatively, approx 0.8% of the world GDP. Imagine the magnanimity of this value by considering that a lot of nations (multiple European nations included) have their entire GDP closer to this value or in multiple cases even lesser.

A lot of fraud detection tools employed in these financial firms involve significant manual support. Data analytics has started making its inroads but with a data deluge at hand and the fraudsters using even more refined ways to fool the system, there is need for more that needs to be done.

How does AI fit in the scheme of these things is what this blog talks about.

To start with, let us first have a look at what an ideal fraud detection solution should offer:

  • Ability to detect outliers/fraud schemes other than those defined by the rules
  • Ability to learn and react in real time
  • Curb false positive instances and hence prevent hiccups to a good customer experience
  • Ability to generate detailed analytics on the outliers/fraud schemes along with the logics to why these flags were raised.

Sounds Good! Now, let us first talk about the case of False Positives.

The Javelin Financial Impact of Fraud 2016 report found that 30% of orders declined as fraud are believed to be legitimate. In fact the report found that the amount lost to false positives outweighs the amount lost to chargebacks by more than 5:1.

False Positives are a majorly common occurrence when it comes to rule based anti-fraud systems. This is because, every now and then a person does make an unusually high purchase or in some cases ends up forgetting the CVV, resulting in transactions getting declined or in worst cases, the card getting blocked altogether.

An ideal system should be able to not just detect an anomaly, but also be able to recognise the probability of false positives such as discussed above. Which means, that a good fraud detection system should not only register the outlier, but also be able to see that the person made the transaction from the usual IP address, the same location, on the same website, for the same item but in a different price range and hence, before going ahead and declining the transaction, should send the person concerned an intimation, asking to check for an additional filter of identity confirmation (for instance, asking the last remembered password to the bank account to confirm if it is the same person).

Thus, the ideal system would be able to identify the presence of an outlier and instead of going ahead and cancelling, it would check for other parameters and install an additional security barrier.

Next in line comes Money Laundering, wherein the key lies in identifying if there is some pattern amongst the specific set of transactions being made by some specific set of people. But, with the billions of transactions being made every minute, how do we come to define the specific- set that needs to be observed.

The United Nations Office on Drugs and Crime (UNODC) conducted a study to determine the magnitude of illicit funds generated by drug trafficking and organised crimes and to investigate to what extent these funds are laundered.  The report estimates that in 2009, criminal proceeds amounted to 3.6% of global GDP, with 2.7%  (or USD 1.6 trillion) being laundered.

The question that thus arises is: while there are so many possibilities, how can one feed multiple conditions into the system especially since  these modalities will keep changing?

Now, that is where AI comes into picture.

 

Understanding the Process

There is literally no dirth of data and though for manual or even for a lot of analytics based systems, it becomes a problem, the Artificial Intelligence based systems benefit from it.

The logic is simple: just like human brain learns on the go as it comes across more and more data points, so does an AI system.  With the usage of Natural Language Processing (NLP) combined with Machine Learning and Deep Learning (ML/DL) algorithms, the AI systems take up cues from what is being registered in the human-speak’; check it with the already inbred rules, compare with the patterns registered and stored over time and then conclude.

This process wherein the system uses self registered patterns to reach to a conclusion is called Unsupervised Learning. With the speed and scale at which fraudsters work at confusing the data, only unsupervised learning can catch up. To give a brief outline of how it works, have a look at the graphic shared below:
Understanding Fraud Detection & Security
*For the specific case of Money Laundering, more than the Deep Learning Aspect, it would be Natural Language Processing at play. NLP uses alogirthms to identify and interpret the human-speak’, thus drawing insights from basic speech or writing. NLP can thus detect and determine connections between a set of people and identify if any false names, aliases or even random transactions to a set of accounts are being used to avoid detection.

 

A Few Concerns- Here and There

The fraudsters, at the end of the day, are humans too! Considering the ways they have been coming up with newer techniques to dodge the computer systems, it is actually a fight of a human brain against a human brain look-alike.

The statement might give the feeling that AI is bound to lose but then, is it not the human brain at work developing the security walls? This brings us to the ‘stage-Next’ advancement being made in the anti-fraud AI tech. Usage of CNN or Convolution Neural Networks can help the AI system process even the most visual cues. Thus, for instance, a security camera feed being integrated in the detection system can help the AI tech to identify if the person using the card at the ATM is actually the person who owns it or not.  Though still in pilot, CNNs, once implemented will take fraud detection to a whole new level of refinement and effectiveness.

 

And to Conclude

We would say that AI anti-fraud tech has come a long way from where it started. From just being a data analytics based business intelligence provider, AI has come to the extent whereby it helped firms like Mastercard reduce their false positives to the tune of 50% while achieving an increased fraud detection by approximately the same percentage.

There are however two key challenges at hand: first , the quality of data that is being fed and the second being, how much investment is justified.

The answer to the first is simple: there is enough data in general concerning the customers and there is enough data for fraud detection and also for false positive. All this is a strong base to start with, and with there being a significant rise in online transactions, the possibility of the data quality improving over time is almost certain.

For the challenge two, i.e. how much is ‘too much’, well, it all depends on how comfortable one is with losing out on the customer’s trust and hence the clientele over time. A basic cost benefit is all that is required.

The answer, however, is simple! Sooner or later, the anti-fraud tech (and more so AI) will have to be implemented. The only matter of choice is – When?

 

(At Racetrack, we understand that how while a banking organisation will be more interested in fraud detection, a credit card firm would be more about credit scoring and then there would be a third firm which would have money laundering as its concern rank 1. Therefore, we offer customised Artificial Intelligence for each business so that they benefit the most.)

  • AI in Hospitality Sector

    AI in Hospitality Sector

    Learn More

  • Loan Approval An AI Powered better Customer Experience

    Loan Approval: An AI Powered better Customer Experience

    Learn More

  • Powering Sales Intelligence in Real Estate

    Powering Sales Intelligence in Real Estate

    Learn More