And now, the rise of the machines — or, at least, machine learning. This movement has been a long time coming, as artificial intelligence (AI) is moving ever-farther into real-world business applications where once it was too expensive (or too conceptual) to embrace.
Through improving technology and decreasing costs, AI and Big Data are now combining to help firms in the financial sector prevent payments fraud.
A recent whitepaper by data science company Feedzai, “The Dawn of Machine Learning for Banking and Payments,” offered up some food for thought (and points to consider) when delving into fraud and risk protection aided by machines.
The Stage was Set with Big Data
It wasn’t too long ago that AI was confined to academia, theory and tinkering from multi-billion dollar companies — firms that could afford to sink millions into its research and development. But now, AI has become more attainable as a tool that companies can tailor to their own operations.
That affordability and ability to scale from smaller firms come against a backdrop in which ever-smaller form factors, increasingly powerful processors, cheaper data storage and the rise of Big Data have all collided to set the stage for machine learning to be fully embraced by the corporate world.
Certainly, the mountains of data are becoming larger by the day. Seven years ago, the total amount of information produced on a global scale passed one zettabyte. The scale shakes out thusly: If a single cup of coffee holds a gigabyte, then the Great Wall of China stores a zettabyte. In just three years, the tally will be 44 zettabytes, or 44 Great Walls of China, as estimated by global market intelligence firm IDC.
The past 15 years have seen technology evolve enough so that this data is not only just attainable and digestible, but can also be used in context and in ways that can truly transform the way companies operate and serve their customers.
Explosive Growth is in the Works
Along with the impressive growth in data created, stored and used on a global scale, so too is AI poised to grow in leaps and bounds. It will create nearly $37 billion in annual revenues for companies of all stripes, sizes and sectors, according to market intelligence firm Tractica.
Within that figure, machine learning is a sector that will see $15.3 billion in revenue in 2019, as noted by BCC Research and cited by business process outsourcing company TeleTech, with an average annual growth rate of 19.7 percent. The savings for U.S. companies could be as high as $60 billion in 2020, Forbes noted. In addition, AI is expected to add $8.3 trillion in economic activity for the U.S. by 2035, according to projections by business management consultancy Accenture.
Continually Adapting Where Old Rules Fail
Well, it is called “machine learning,” after all. The “learning” succeeds where traditional computer programming fails. That’s because traditional computer programs are governed by rules that are often updated in a batched mode.
Machine learning, by contrast, updates information continuously and algorithms create “models” that take in data inputs — now and in the future — and translate them into useful output. The system is trained and then consistently scores real data, often in real time.
Consider a financial institution processing credit card information. The transaction data is passed to the machine learning system as soon as it is entered at the terminal or point of sale, and the system then analyzes the transaction against the system on which it has been trained. The historical data offers a way to glean what “normal” behavior of a transaction looks like.
Finding, Learning About and Explaining Data — but the Machine Needs Permission
It’s no secret that traditional methods of fraud detection are inefficient at best. Manual processes dominate fraud prevention efforts more than one might think, accounting for as much as 25 percent of the total cost of fraud prevention, as a LexisNexis Risk Solutions study on the “true cost of fraud” found.
Where manual processes abound, though, it is likely costs can be reduced. Machine learning can find patterns not readily apparent to human observers, and such platforms can also identify deeply hidden threats from malware while weeding out false positives.
To combat a payments fraud adversary that is evermore fluid with bad actors’ tactics and operates in a card-not-present (CNP) world, the machine deployed by a financial institution must be able to “explain” what it is doing, Feedzai said. The “learning” should result in explaining the reasoning so the logic behind the decisions is transparent and meets compliance needs.
Of course, the technology put in place must be robust, agile and flexible enough to handle change and scale as the very organizations deploying the technology change and scale. Feedzai noted machine learning systems must be able to “see” and extract the appropriate data for analysis — and handle a ton of volume in the process.
The system must also be able to adapt as the surrounding world changes, too. After all, data is not static, nor is the type of data that is going to be analyzed ever truly static. The system must be agile to act on its findings with the ability, for example, to refer suspicious transactions to human analysis for additional review.
To Download: The Dawn of Machine Learning For Banking And Payments, please fill out the form below:
Perhaps it goes without saying that in a world rife with fraud that can occur in mere seconds, speed is of the essence.