You are currently on IBM Systems Media’s archival website. Click here to view our new website.

POWER > Business Strategy > Competitive Advantage

Financial Institutions Use AI to Improve Operations

AI and Banking
Illustration by Wayne Mills

Most artificial intelligence (AI) headlines these days tend toward the sci-fi, with social-media companies and governments using it to control our minds and permanently eradicate any notion of privacy. Less common are the examples of AI being used in the real world to simply improve business processes, increase profitability and improve customer service.

One such industry that’s openly adopting the technology is banking and finance. Businesses such as PayPal and a large European financial institution, for example, are using AI to reduce instances of fraud, thereby lessening associated costs and increasing customer loyalty. Others, such as ANZ Bank, are using the technology to improve credit approval times and reduce credit defaults.

AI is quickly becoming an everyday part of conducting business. As examples in the banking and finance industry demonstrate, AI can—when imaginatively and properly implemented—have a positive outcome for both companies and the customers they serve.

Fraud in Disguise

Fraud, of course, impacts many industries, but perhaps nowhere as pervasively as in banking and finance. This can include stolen credit and debit cards, pilfered card numbers and ATM scammers. Although companies involved in banking and finance have been at the forefront of reducing fraud, their efforts were somewhat stymied by a lack of detailed analytics and insight that could help them more easily spot and reduce instances of sham transactions, while also reducing false positives and negatives.

As Scott Soutter, portfolio offering manager for PowerAI, points out, “You can get in trouble with false positives and negatives. If you clear a transaction that, in hindsight, is obviously fraudulent, you lose the money associated with that transaction. And if you block somebody from doing something that is actually above board, you could potentially lose a customer. These are both costly errors.”

That’s in part why companies such as PayPal and other large financial services companies decided to employ deep learning as a method by which they could speed up and more accurately tackle this issue. The large European-based financial services company’s goal, for example, was—using accelerated IBM Power Systems*, Power AI and the open-source deep-learning tool Theano—to combine existing sets of analytical models with deep learning to find additional hints about which transactions are or aren’t fraudulent.

It could then more quickly determine whether to halt or allow specific transactions using information about a particular customer’s typical behavior. PayPal took a similar approach by also using accelerated Power Systems and PowerAI to institute deep learning to prevent fraud by quickly targeting extra-large datasets, something that wouldn’t have been feasible using typical programmatic detection methods.

China Industrial Bank (CIB)employed deep learning to help prevent fraud and crime at unattended ATM machines by using facial recognition and other detection techniques to ascertain if someone is purposefully hiding their facial features, which might indicate that a potential robbery is about to take place.

“People were making unauthorized withdrawals from the machines,” Soutter says. “So CIB figured out that a lot of these people were hiding the features of the face, by, for example, pulling their hats down low and wearing the medical masks that are so common in parts of Asia—characteristics like those, where people were occluding parts of their features as they go to an ATM. The challenge is being able to describe what that looks like and actually train a model to look in real time at what’s happening in front of the camera that sits inside of that ATM and then be able to immediately in stop something that’s happening.”

To that end, CIB created a model using the PowerAI Vision application wherein they were able to see the characteristics of somebody hiding their face from the camera and either slow down or stop transactions depending on those characteristics. If the bank slows a transaction of a legitimate customer, the customer will typically wait for the completion of that transaction. If, however, it slows the interaction of somebody who may be attempting a fraudulent transaction, that person would generally abandon the activity.

Credit Where Credit’s Due

Most companies have to deal with customer service in some way, whether it’s business to business or business to consumer, and businesses involved in banking and finance are keenly aware of how vital this is to customer retention.

That’s why companies such as ANZ Bank and Caixa Geral de Depoósitos France (GDP France) have begun using AI to streamline their credit approval processes. Using Power AI and focusing on data preparation and transformation, for example, ANZ can draw from multiple organizational data stores and create a space where data scientists can address multiple new deep-learning models.

The first of these models focused on reducing credit approval turn-around time, allowing it to more quickly address customer requests while also implementing a more accurate predictive model that helps reduce the potential for credit default. The second model focuses on upselling to existing clients, thereby improving its sales and marketing efforts to increase current account profitability.

“We’re trying to help our customers take advantage of the tools and the capabilities by simplifying the how of implementing these types of deep-learning and machine-learning AI solutions."
–Scott Soutter, portfolio offering manager for PowerAI

“Being able to understand patterns in what customers are doing and how they’re changing, the way that they want to interact with products, is a very common marketing use case,” Soutter notes. “So getting to a level where you understand how whole communities are operating or taking and reusing that to a very personalized experience with an individual is vital. Success in this area is about knowing how to address that individualization, and that’s where AI can be of great use.”

This is also true in the case of GDP France. Its inefficient paper-based credit scoring delayed personal-loan applications and required customers to revisit their branches for final decisions. This raised the risk they would choose another bank—and result in other business losses, such as savings and checking accounts and the number of credit-card holders. This was unacceptable, especially to the sales team.

Recognizing that this was an issue, GDP France decided to use machine learning in a Power Systems hybrid-cloud environment. Its ultimate goal was twofold: giving credit analysts a tool to model and maintain multiple risk models for various types of customers; and using these models to perform real-time scoring for loan applicants. This would essentially eliminate the need for multiple branch visits, and thereby reduce the likelihood of customer defections.

“This type of solution has to be accurate and it has to lead to a successful outcome for the business,” Soutter remarks. “So if you deploy it properly, you can elevate your brand and improve the way you interact and reinforce that experience for your customers.”

Dollars and Cents

As some players in the banking and finance industry have proved, AI can be used for myriad purposes, whether for fraud detection and prevention or improving customer service. To them, this comes down to bottom-line dollars and cents.

The same can also apply to most any other industry, be it manufacturing, logistics or airline operations. And Soutter says these users don’t have to be enterprise size—merely interested in how they can improve processes for both themselves and the organizations and people they deal with. As he further notes, “We’re trying to help our customers take advantage of the tools and the capabilities by simplifying the how of implementing these types of deep-learning and machine-learning AI solutions. They don’t have to become core experts in the underlying software required to generate a neural network. They can instead use technology like ours—including integrated software and servers—to develop their own models to conduct business more efficiently. After all, they’re in the business of business, and not the business of developing AI solutions from scratch.”

Jim Utsler, IBM Systems Magazine senior writer, has been covering the technology field for more than a decade. Jim can be reached at



2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.


5 IBM i Architectural Principles Drive Business Innovation


How the API economy is changing business

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store