AI on IBM Z & LinuxONE - Group home

AI solutions on IBM Z and LinuxONE can help curtail credit card fraud before it happens

By Ismath Mohideen posted Mon May 17, 2021 01:02 PM


What is the problem with credit fraud?

You get a text or open your credit card app and see it: an unknown charge. Maybe it’s a few small charges that fly under the radar, or a large charge in a distant location. Your credit card information was violated by an assailant trying to steal your money. Frustrated, you contact your credit card company to solve the problem, only to discover it will take 45 days to conduct a report. Your frustration grows, and so does the frustration of credit card companies as they encounter this problem on a widespread basis.

Credit fraud, including credit card fraud and identity theft, is a growing international problem affecting banks and their customers. 40% of financial service firms have encountered an increase in fraudulent activity since the pandemic began, including unauthorized charges and identity theft1. As customers use ubiquitous online access to banking and financial services especially in newly mobile and connected economies, the risk and notoriety of credit fraud grows. This growth and the volume and velocity of transactions and data generated make manually identifying and rectifying credit fraud an impossible task. Additionally, banks must adhere to regulatory compliance when addressing fraud verification and customer identity management, and they face steep fines for noncompliance. Without artificial intelligence tools, credit fraud becomes expensive and daunting, and rules-based methods cannot keep up with continuously morphing fraud schemes.

How much is this costing the industry?

Banks are subject to laws and regulations over fraud protection, data privacy and customer data protections. In 2020, global financial institutions faced over $10 billion in penalties related to anti-money laundering (AML), know-your-customer (KYC) noncompliance, data privacy and derivative regulations, and those headquartered in the US accounted for nearly 75% of the total fines2. Furthermore, global credit card fraud is expected to reach $35 billion by 20253.

Credit card fraud is usually detected by implementing rules around known fraud patterns or anomaly detection that recognizes unusual activity, such as knowing historical charge amounts or regions where charges typically happen. Anything deviating from historical norms is flagged as possible fraud and may result in declined charges. These rules are typically static and do not change unless they are identified after manual, time consuming analysis. The rules and patterns are slow and difficult to evolve as fraudulent transactions become more sophisticated. The result? Missed opportunities to correctly capture fraud and incorrectly denying valid transactions, which increases costs and decreases revenue. Artificial intelligence (AI) helps fight credit card fraud by searching for and acting upon learned suspicious activities before they become major problems.

The largest banks in the world rely on the IBM Z and LinuxONE platforms to process transactions. Using AI at the exact same time the credit card is in use, referred to as the transaction level, banks can stop fraudulent transactions using fast machine learning-based analyses. Transaction-level AI can halt transactions before completion and data scientists can modify the rules around fraud detection. This reduces the need for human intervention after fraudulent transactions have occurred, resulting in lower overhead costs, reduced risk of fines, and better customer satisfaction. Banks can then focus on improving customer relationships, and by using AI solutions on the IBM Z and LinuxONE platforms, banks can begin their journey to AI.

How does artificial intelligence help in tackling credit fraud?

AI techniques provide for intelligent modeling capabilities that can learn behavior patterns from existing data sources. AI models can keep learning to adapt to trends or shifts in customer behavior to stay ahead of potential fraud. This allows a financial institution to effectively update their fraud detection approach based on real feedback from data, resulting in significant revenue potential. In fact, a McKinsey study illustrates that AI can potentially reveal $1 trillion of incremental annual value for banks4.

Different classes of AI can be applied to problems like fraud detection. Machine learning refers to learning patterns in existing datasets and has broad penetration in predictive modeling, but deep learning adoption is growing. Deep learning utilizes neural networks to identify complex relationships between data, learning patterns in both structured and unstructured data. Well known deep learning programs feature image and speech recognition. Going further into deep learning, models such as long short-term memory (LSTMs) are proficient at analyzing time sequence problems, such as recent trends in client behavior. These applications are capable of identifying the dynamic changes in fraud tactics better than traditional business rules. Different types of AI serve different purposes when combating credit fraud, but running these applications close to where the transactions run and the data resides on the IBM Z and LinuxONE platforms provide the groundwork for uncovering fraud detection insights in near real-time.

Why are the IBM Z and IBM LinuxONE valuable enterprise platforms to combat credit fraud?

Performing AI analysis in a security-rich environment where transactions are generated, documented and reside is crucial for combating credit fraud at scale and with low latency. The IBM Z and LinuxONE enterprise platforms are renowned for high-volume, high-throughput transaction processing for mission critical applications, with industry leading encryption capabilities, security for data at rest and in flight, and optimized availability for continuous operations.

While you can train models anywhere, the data gravity (large data sets attracting applications and services close to where the data resides) of IBM Z and LinuxONE make it an ideal environment to conduct AI analysis or inferencing directly on the platform. This is because performance, scale, and latency are improved when models are deployed close to the data. Typical AI analysis requires transferring data off platform, resulting in increased latency, security risks from moving data, costs of copying data, and added time to finalize an analysis. AI inferencing on IBM Z, where transaction level data resides, reduces or eliminates the need for copying and transferring data sets, reducing model analysis time, the likelihood of security breaches when transferring data, and the cost required to copy and transfer data.

Battling fraud and leveraging AI requires credit card companies to ensure that systems are continuously running with minimal downtime. According to the 2020 Hourly Cost of Downtime survey by Information Technology Intelligence Consulting (ITIC), the cost of one hour of downtime exceeds $300,000 for 88% of firms surveyed, and even more for large financial institutions5. ITIC ranks the IBM Z platform with the highest reliability and availability ranking among mainstream server platforms with up to 99.99999% uptime, for reduced downtime costs while running credit fraud AI models in an uninterrupted fashion5.

Check out our AI Content Solution to learn more, and talk to us about our Sponsor User program.

We are enabling financial institutions to fight credit fraud by optimizing the deployment of AI workloads and inferencing on IBM Z and IBM LinuxONE with innovative hardware and software solutions. Learn about our solutions to operationalize your AI models, including IBM Cloud Pak for Data and IBM Watson Machine Learning for z/OS for flexibility and choice in hybrid cloud environments.

Send us an e-mail or fill out this form for our AI sponsor user program. Let’s have a conversation about how our solutions can be tailored to reach your goals.


1: PaymentsJournal, AI Fights Fraud: How the Use of AI Technologies in Banking Forges the Fight against Fraudsters

2: Global Financial Institution Fines for AML, Data Privacy and MiFID Rise 26% in 2020

3: Nilson Report, Issue 1187, Dec. 2020

4: McKinsey & Company – AI-bank of the future: Can banks meet the AI challenge? Sept. 2020

5: ITIC 2020 Global Server Hardware, Serer OS Reliability Report, April 2020