American Express has been a leader in using artificial intelligence and cognitive technologies for years. As a global financial services institution with a loyal customer base, keeping their customer’s accounts safe and secure has always been top priority. Being able to spot and stop fraud has been a priority at American Express for quite some time. In recent years, the use of artificial intelligence has helped the company really advance their fraud detection efforts.
Rajat Jain, Global Head for Identity and Authentication Strategy at American Express, was recently on the AI Today podcast. He shared how machine learning can be applied successfully to detect and catch fraud, how AI and machine learning are being used for the overall customer experience at American Express, and some surprising insights about AI adoption. He’s also sharing his insights at the upcoming Data for AI Week virtual conference. In an interview for this article, he shares some of his insights into how American Express is applying AI and machine learning (ML), staying at the forefront with fraud detection, and some surprising insights into how AI has been applied at Amex.
How long has American Express been applying AI and ML in its various processes?
Rajat Jain: Early on, American Express’ leadership recognized the need for and value of data analytics and technology, which drove our machine learning transformation across risk, marketing and servicing. In 2010, we began researching machine learning techniques and assessing their potential in core business processes, including credit risk analysis and fraud detection. In 2014, we implemented the first large scale application of machine learning models for fraud detection and we saw an immediate 30% improvement in detection over the previous non-machine learning models.
Our relentless focus on eliminating fraud has resulted in Amex maintaining the lowest fraud rates in the industry for thirteen years in a row, according to The Nilson Report – in fact, we see losses at half the rate of other major networks.
How is American Express using advanced machine learning techniques for fraud detection?
Rajat Jain: We seek to use the latest and most sophisticated machine learning techniques to protect our card members and merchants from fraud. Our machine learning algorithms monitor in real-time every single American Express transaction around the world — that’s more than $1.2 trillion annually, and we generate a fraud decision in mere milliseconds. One of the techniques we use to monitor fraud is Sequential RNN. Data is analyzed sequentially to understand the relationship between transactions to more quickly identify spend that doesn’t make sense or is what we call “out of pattern.” Simplistically speaking, if a customer makes a coffee purchase in New York at 10:00AM and 10:05AM buys a tank of gas outside Los Angeles, we’re going to know right away that the card has been compromised.
How is AI and ML different from previous approaches for fraud and risk management?
Rajat Jain: For decades, financial services companies have been among the earliest adopters of the latest analytical methods, staying one step ahead of fraudsters in order to protect their customers’ personal and account information. The finance industry is still on the cutting-edge, and in my opinion, no single advancement has shifted the success rate of stopping fraud more than machine learning. Machine Learning has three key advantages over previous logistical regression models, they are:
- Capture non-linear trends and interactions between variables more effectively resulting in greater accuracy
- Rapid deployment of a single global modeling solution enabling agility and capture of geographically migrating trends
- Releases team bandwidth which can be invested in advancing data science
How is AI and ML being used to enhance the overall customer experience at American Express?
Rajat Jain: Machine learning innovation at Amex is leading the optimization of the most non-technical application: world-class customer service. Having our Card Members’ backs is our top priority and keeping our fraud rates low is key to achieving this goal. Hundreds-of-billions of data points and billions of decisions every year all feed into a system that provides consumers and businesses with what matters most in today’s world – safe ways to pay, faster decisions, real-time customer communications, and world-class fraud protection and servicing.
How have you seen the use of AI and ML impact the overall customer experience?
Rajat Jain: It’s essential that we connect all the dots across our data, so we can be unmatched in delivering the world’s best customer experience every day. Today, that means showing our customers we know them, understand their needs, and ultimately have their backs in every digital interaction we have with them. As an example, we use machine learning to detect and then provide real-time digital notifications of fraudulent activity to our card members who sign-up for them. We have made it easier than ever for our card members to monitor their accounts for fraudulent activity through these alerts that are delivered in real-time via email, text message and mobile app push notification.
What are some interesting or surprising insights you can share about American Express’s use of AI?
Rajat Jain: American Express’ data science team is comprised of MS and PhD data scientists who are never satisfied with their status quo; they are constantly upgrading their skills. In a field that’s constantly evolving, American Express recognizes the only way to be successful is continuous learning and experimentation in emerging techniques. The team regularly evaluates the strengths and weaknesses of our models and finds ways to innovate on behalf of customers. In fact, next month, we’re launching the Generation X Fraud Model, which will feature our latest innovations to capture even more fraud.
What do you see as the major limitations in applying AI and ML to bank processes?
Rajat Jain: Everything we do is in service of the customer and having their backs. For us, this means the limitations in applying machine learning to bank processes really come down to whether a new analytical technique will actually help us advance our mission of delivering the world’s best customer service every day. It’s easy to get excited about the latest data science research and advancements, but those are theoretical and academic studies. We operate in the real world where the decisions we make impact the safety of our card member’s accounts, so when we apply a new type of AI technique we know that it can be implemented into production and deliver the results that our customers expect.
Can you share with us what data considerations need to be taken at Amex when building ML models?
Rajat Jain: We always say that not all machine learning models are created equal and one of the things that set our model apart is the data we use in it. American Express has a global presence across the entire payments chain, since we operate as a card issuer, merchant acquirer and a network. This “Closed Loop Network” provides us with an advantage in fighting fraud by combining our vast data set with our highly skilled and trained subject matter experts and cutting-edge machine learning algorithm. We have visibility into a tremendous amount of data from both merchants and card members, which enables us to act more quickly than other networks and issuers to identify and stop instances of fraud before they rise to substantial levels.
How does American Express ensure customer privacy and data privacy when building ML models?
Rajat Jain: American Express recognizes the importance of maintaining consumer trust and has a strong, long-standing commitment to privacy and data security. We protect personal data in accordance with our Data Protection and Privacy Principles and Binding Corporate Rules, as well as applicable laws, our contracts and other internal policies.
What AI technologies are you most looking forward to in the coming years?
Rajat Jain: Broadly, I see that we are just beginning to scratch the surface when it comes to enabling predictions on sequential data sets. Artificial intelligence has yet to 100% effectively stitch together a sequence of events in this way, so I’m excited to see how progress in this area will enhance the customer experience. As for how we will combat fraud in the coming years, we’ll continue doubling down on our offensive and defensive controls to counter new threats by continuing to evaluate new techniques and apply the ones that deliver better fraud protection.