using ai and big data in fraud prevention

Feeding Big Data to AI Helps Banks Fight Cybercrime

It might look like AIs are something that will only become reality in the future but the truth is, AIs are already here. All around us. 

An AI helps us search the web much more efficiently. We call it Google. There is also one that listens to our voice and helps us shop better. We call it Alexa. There is even one that helps us protect us in case of a car crash. We call it Air Bags.

Each of these AIs used regularly today is a program that thinks and with each web search, with each instance of voice shopping and driving, these AIs are getting more and more data that helps them learn better.

The very same approach - feeding AIs vast amounts of data, i.e. Big Data, helps banks discover fraud more efficiently.

Physicists ask what kind of place this universe is and seek to characterize its behavior systematically. Biologists ask what it means for a physical system to be living. We in AI wonder what kind of information-processing system can ask such questions.

Avron Barr and Edward Feigenbaum

And banks certainly have a lot of data. They are one of the biggest contributors to the amount of data being created in the world today, an amount constantly on the rise. According to IDC, by 2020 there will be 44 zettabytes of data in the world. This kind of growth makes AIs scale rapidly in their accuracy when predicting fraud. The growth of cybercrime makes the use of AIs a necessity.

According to a McAfee survey, it is now estimated that cybercrime is costing the global economy 400 billion dollars per year. This is why the use of AIs in banking is becoming diverse, as it is also used in machine learning, predictive modeling, data mining and advanced statistical modeling.


"In general, AI includes statistical methods, computational intelligence and symbolic methods for problem solving and learning", explains Nataša Benčić, Senior Quantitative Analyst at Mercury Processing Services International. 

"We use the AI methods to recognize unusual behavioral patterns in the set of all transactions that come to our authorization systems."

This skilled mathematician, who works in Mercury Processing Services International's Fraud Intelligence and Product Management, developed the fraud scorecards for the in-house Lynx Fraud Management System which led to signing a partnership agreement for the model development between Mercury Processing Services International and SAS Institute Inc., world leader in analytics.


"As suspicious (high risk) authorizations are recognized by the models, the fraud prevention and detection process will be optimized by helping choose only a subset of all authorizations that has to be rejected in the fraud prevention process, or analyzed further more in the fraud detection process", Nataša clarifies how machine learning helps fraud detection and prevention strategies, also giving more insight into the applications and benefits of predictive modeling.

Top Uses of AI and Machine Learning-01_a.png

From Healthcare to Banking

"As predictive modeling uses past behavior to predict the future outcome given a set of input data, it can be applied to various fields – customer attrition, customer relationship management, cross-selling, up-selling, insurance, healthcare, finance and banking risk management, telecommunications, travel, pharmaceuticals, collections, fraud prevention and detection," she continues.

"The goal is always to optimize a process, to maximize profit and minimize costs using the outputs of a statistical model for decision management. Of course, the predictivity of a model should be checked regularly and corresponding actions should be performed in order to assure that the right decisions will be executed."

Data mining and advanced statistical modeling are widely used in Mercury Processing Services International's business. Nataša gives examples: "Most of the models we've developed were for banks' risk management purposes – application and behavioral scoring models, where client's performance on active accounts is tracked. The method used is logistic regression. The results are presented in a scorecard form which is easily explained to end users and quite easy to implement in the system. Some of the models are also already implemented in our systems. Besides that, we've developed attrition models and collection models for some of our clients. Also, our Lynx Fraud Management System was enhanced with a scoring module that improves the fraud detection capabilities mainly by improving FDR and decreasing the number of false positives."

The Future Is Now

  • Data production will be 44 times greater in 2020 than it was in 2009.
  • 6 million developers worldwide are currently working on big data and advanced analytics.
  • The worldwide business intelligence and analytics market will be worth $18.3 billion by the end of 2017.
  • 1.7 megabytes of data will be created every second, for every person on Earth by 2020.
  • 100% of IoT initiatives will be supported by AI capabilities by 2018.
  • 75% of enterprises and software vendors will include cognitive/AI functionality in one or more of their applications or services by 2018.
  • 40% of mobile interactions will be managed by smart agents by 2020.
  • Total worldwide spending on AI by 2020 will amount to $47 billion.
Sources: Wikibon, Evans Data Corporation, Gartner, Helios Solutions, Tech Pro Research