As one of the so-called “big four” U.S. banks, Chase needs little in the way of introduction. And like many age-old institutions, including its direct rivals, the New York-based financial powerhouse has had to move with the times, with Chase now investing more than $11 billion each year on the technology side of its business. This includes software development, cybersecurity, and — increasingly — artificial intelligence (AI) and machine learning (ML).
Talking at VB Transform today, Sandra Nudelman, chief data and analytics officer at Chase for the past two years, outlined some of the main ways it’s harnessing AI and ML across its business, including helping it to streamline internal processes such as managing PPP applications, improving their marketing efforts, increase credit lines, and preventing fraud.
In response to the COVID-19 crisis, the U.S. government launched the Paycheck Protection Program (PPP) a couple of months back to ensure money continues to roll into the workforce — this, in turn, led to significant paperwork for banks, which have had to deal with a mountain of applications. The Small Business Administration (SBA) reportedly had to process 75 years’ worth of loan applications in just two months, which gives some idea as to the scale of this undertaking.
Faced with such an unprecedented challenge, one that affected the lives and livelihoods of literally millions of Americans, Chase had to come up with a way of classifying documents that its customers were uploading as part of the PPP application process. It did so with a view toward helping its business banking division and underwriters wade through as many applications as possible.
“They needed a way to understand what documents our customers were uploading, which we hadn’t yet tagged every single document as part of our workflow,” Nudelman explained. “So instead, after the fact, we worked with the people building the process and technology to use natural language processing (NLP) to ensure the documents that have been uploaded were tagged appropriately, which helps the underwriters’ ability to process those applications, getting customers their loans faster.”
It’s worth noting here that the underlying technology it used was already in place as part of its broader machine learning platform — it just needed a little fine-tuning.
“This was just about retraining and tuning [the machine learning model] for a new set of documents and a new purpose, which happened — I think — over the course of a weekend,” Nudelman added.
The issue of gender bias in algorithms hit the headlines last year after Goldman Sachs was accused of giving a man a credit limit that was 20 times higher than his wife. Thus, any discussion around AI’s role in decisions related to credit worthiness is a potential hot potato for banks.
According to Nudelman, Chase has started dipping its toes in the waters around this area, but is taking things slowly to avoid scenarios that may lead to bad decisions that can’t be explained.
“We want to be very, very careful, that anything we do in this space is fair to all of our customers, and that we actually understand what the models are doing as we work towards them,” Nudelman said. “So we’re inching our way in, and focusing on things that are a little less risky. For example, things like proactive credit line increases are a place that regulators think we can play a little bit more into, as opposed to things that might be derogatory decisions for customers.”
AI in marketing
Last year, Chase inked a five-year deal with Persado, a platform that leverages AI to personalize marketing messages, with Chase chief marketing officer Kristin Lemkau saying at the time that “machine learning is the path to more humanity in marketing.” Building on this notion, Nudelman said that AI and ML enable them to take a more measured approach to their marketing efforts, with algorithms used to interpret data and other feedback signals to select the best messaging for each customer. This, according to Chase, has led to a nearly fivefold increase in engagement with its campaigns.
“What we really try to do in marketing is to connect with the customer, and in the past we had a creative come up with the content that we thought would resonate best — we [would] take a bet and we get it out the door,” Nudelman said. “Now what we try to do is we build systems that are agile and take multiple different feeds, we experiment, and we enable the machine to look at how customers are responding to that creative and to pick the winner itself.”
According to Nudelman, this is less about issuing overly familiar messaging of the type that can come across as creepy, and more about delivering campaigns that speak to each customer’s personal circumstances — in other words, it strives to acknowledge that no two customers are the same.
“So if you save [money], and we think you could be saving a little bit more based on the spending patterns that we see in your account, we might put an advertisement up that says, ‘you could be saving $70 a month more, click here to auto-save that amount,” Nudelman said.
The fraud factor
AI and automation have emerged as integral tools for cybersecurity teams across industries, simply because the sheer number of external (and internal) threats far exceeds the capacity of humans alone to tackle. As such, Chase is levering both supervised and unsupervised ML to spot known threats based on previously identified patterns, as well as potential new threats that it has not yet previously logged. Ultimately, it’s about preventing fraud from happening in the first place, rather than picking up the pieces afterwards.
“Going after and identifying fraud proactively to prevent it is something that we’ve put a lot of effort in,” Nudelman said. “We’ve been able to decrease fraud [of] well over $100 million a year using these efforts.”