Outlier Briefing: The emerging use cases for AI in financial services with Emerj’s Dan Faggella

Welcome to Tearsheet’s Outlier Podcast. This subscriber-only podcast is exclusive for our Outlier members. We go deeper with subject matter experts, to take actionable steps that can impact your business and market.

Today’s guest is Dan Faggella, a global expert on artificial intelligence and its use cases for financial services. Dan’s the founder and head of research at Emerj, an AI research and advisory company. Dan takes a practical approach when it comes to discussing AI’s potential in financial services.

The adoption challenges for bank take up of AI are real — Dan polled them and has built a framework around ROI on investments in artificial intelligence and how to overcome implementation hurdles. We talk about concrete use cases for AI in the front office and where we’re headed when it comes to cross-selling or customer acquisition. You can read Dan’s take on the critical capabilities — the prerequisites — needed for AI deployment.

Subscribe: Apple Podcasts I SoundCloud I Spotify I Google Podcasts
The following excerpts were edited for clarity.

Takeaways

• The financial industry’s dance with AI: “Artificial intelligence has a somewhat storied history in financial services. Before Google and Facebook and everyone moved to Silicon Valley when they got an awesome AI degree from MIT or Carnegie Mellon, they all moved to New York. That was the original place where AI talent would cluster after they graduated.”
• ROI on AI: “[When we work with a client], we look at two main things. The first is the evidence of ROI. We do hard work, primary research to figure out the hard return on the investment in these technologies. The second is which of these technologies are inevitable? That’s to say the AI researchers, the startups, the buyers of tech in banking are all in agreement that this is a no-turning-back, going-to-change-the-way-we-do-business trend. There are plenty of those in finance but we’re in very early days.”
• Clearing the AI hurdles: “There is a lot to get AI off the ground. Among those ingredients is some semblance of in-house, data science talent. The thought that we can be ignorant of the potential and challenges of AI and a vendor could come in and make AI work — plug AI in — it’s an exceedingly naive notion.”
• AI versus automation: “One of our research advisors is a former head of AI at HSBC. He looks at where IT and automation budgets are headed. Should AI be included under automation? Probably not, but it often gets classed under automation.”
• Bank culture: “After 2008, banks are risk-focused. With the fear of regulatory and compliance, there’s an emphasis on playing defense and playing it safe. So, where attention goes, energy flows. Compliance and fraud will get attention for AI before nifty chatbots.”
• AI skills: “There are two skillsets. One of the skillsets is getting artificial intelligence to produce a result reliably. You can learn this in school. You go to Stanford or Carnegie Mellon, you get a degree. You get enough data and context with business people, you can train an algorithm to give you an output that’s reasonably effective. Enterprise search, for example. If this utterly fails 20 percent of the time, it’s fine for applications we use inside the bank.”
• Customer service apps that use AI: “For apps that directly interface with customers, we now need something much less than an 80 percent working rate. What this implies is that we need a really snuggly bounded reality as to what kinds of tickets and inquiries we can handle with this algorithm and which ones we can’t.”
• AI talent in banking: “We need really fast methods of routing these messages to a human being as soon as we get to the end of the rope of what our AI program is capable doing. This flywheel of skills is much more rare than having a PhD from an Ivy League school.”
• Lens of incentive: “For any given category of AI innovation or investments, there are some that we want to exaggerate and emphasize as a bank. Because these make us look better to customers and shareholders. There are other ones that we want to downplay or hide because they don’t make us look good — or make us look bad — to our stakeholders.”

Listen to the whole briefing

Dan’s background

I’m the head of research and CEO at Emerj Artificial Intelligence Research. Think of us as a boutique Forrester or Gartner except that we only have one focus: the ROI of AI in financial services. We poll the AI landscape in fintech, banking, insurance and wealth management — where everyone is spending their money.

I didn’t study computer science — I studied how humans learn. At University of Pennsylvania, I studied skill acquisition and development. It was a psychology/cognitive science discipline. The folks in the computer science world were saying that the neuroscience I was learning was being complemented by what they were doing with machine learning. This was 2011, when machine learning was beginning to get hot in academics. I became fascinated by it.

AI and financial services

Artificial intelligence has a somewhat storied history in financial services. Before Google and Facebook and everyone moved to Silicon Valley when they got an awesome AI degree from MIT or Carnegie Mellon, they all moved to New York. That was the original place where AI talent would cluster after they graduated.

In addition to that, financial services spends on hardcore R&D to get an advantage. Finance is a big industry that spends on pilot projects. They are learning more from pilots pound for pound than other sectors out there.

AI in finance: overrated or underrated?

It’s probably overrated in most cases. That’s not to say that AI won’t be changing a lot of the core processes in finance. When we work with a client, we look at two main things. The first is the evidence of ROI. We do hard work, primary research to figure out the hard return on the investment in these technologies.

The second is which of these technologies are inevitable? That’s to say the AI researchers, the startups, the buyers of tech in banking are all in agreement that this is a no-turning-back, going-to-change-the-way-we-do-business trend. There are plenty of those in finance but we’re in very early days.

AI landscape today

There are two categories. The first is the very large FIs like JPMorgan and Citi. Some of these are naturally more tech savvy than others. Capital One, for example, has more of a technical bent than HSBC. When we look at these largest firms, they have applications for AI.

When we look at the mid market, like a Citizens Bank, their AI activity is middling at best. Once you dip below the top 10 US banks, you’ll be scrapping to find tiny bits of AI.

On the fintech side, they have a lot of venture money and they have digital, data, and infrastructure in their blood.

We have a great article called Critical Capabilities to get AI off the ground. There is a lot to get AI off the ground. Among those ingredients is some semblance of in-house, data science talent. The thought that we can be ignorant of the potential and challenges of AI and a vendor could come in and make AI work — plug AI in — it’s an exceedingly naive notion.

People are just uneducated on how AI works. A lot of our work is with firms that want to transform. What they don’t want to do is blow $6 million on pilots in 16 different directions.

There needs to be a stomach for R&D. People might invest in just cleaning their data for six to eight months in order to get the next big test round up to snuff. Not all these projects will work out. There are some applications that have a better hit rate — others are speculative. We never really know if a pilot will work in our IT environment, if we have the in-house chops to get it done, or if our data will cut the mustard to deliver the results we expect it to have.

AI vendors know this and are, by and large, selling exclusively to the enterprise.

Today’s use cases for AI in financial services

The preponderance of applications are related to risk: fraud is a nice low-hanging fruit area. That’s true with Visa and Mastercard around payments, but the same is true for bank customers trying to open a bank account.

Cybersecurity is another big domain as well. This is an inevitable business function undergoing change. Lending is another domain — whether to say yay or nay to an application and what kind of rate to give the applicant. This is an ML-informed decision.

On the customer service side, there are applications around chat, IVR, and phone. There are small applications in marketing and sales — frankly, not much.

Risk vs. customer facing apps

There are two main factors. One of our research advisors is a former head of AI at HSBC. He looks at where IT and automation budgets are headed. Should AI be included under automation? Not precisely. It isn’t the same as RPA, but it often gets classed under automation.

Another thing to look at is culture. After 2008, banks are risk-focused. With the fear of regulatory and compliance, there’s an emphasis on playing defense and playing it safe. So, where attention goes, energy flows. Compliance and fraud will get budget before nifty chatbots.

AI talent and skills in financial services

There’s another thing that almost never gets talked about and that’s the issue of skill. There are two skillsets. One is the skillsets is getting artificial intelligence to produce a result reliably. You can learn this in school. You go to Stanford or Carnegie Mellon, you get a degree. You get enough data and context with business people, you can train an algorithm to give you an output that’s reasonably effective. Enterprise search, for example. If this utterly fails 20 percent of the time, it’s fine for applications we use inside the bank.

For apps that directly interface with customers, we now need something much less than an 80 percent working rate. What this implies is that we need really snuggly bounded reality as to what kinds of tickets and inquiries we can handle with this algorithm and which ones we can’t. We need really fast methods of routing these messages to a human being as soon as we get to the end of the rope of what our AI program is capable doing. And we need a swift iteration schedule to determine how well we’re understanding different categories of inquiries and where we’re dropping the ball on customer relationships.

This flywheel of skills is much more rare than having a PhD from an Ivy League school. And almost everyone with this skillset is in Silicon Valley. So, it will take banking world quite a while to get these people with these skills. It makes the customer service/marketing and sales side of things very hard in big, clunky established financial institutions.

Getting budget for projects

Most C-suite folks still see AI as IT. A lot of bankers will think that AI means chatbots in banking based on a presumption that chatbots are very popular.

We talk about the lens of incentives: for any given category of AI innovation or investments, there are some that we want to exaggerate and emphasize as a bank. Because these make us look better to customers and shareholders. There are other ones that we want to downplay or hide because they don’t make us look good — or make us look bad — to our stakeholders.

Say, Citi invests tremendous amounts of money in anti-money laundering to prevent terrorists from buying bombs or shipping stolen artwork. Is Citi cooler when it does a press release about that? Will anyone give me props for that? Or, do people just get nervous that terrorists were buying bombs with money funded through my institution?

For that reason, fun, happy, neato customer facing stuff gets an inordinate percent of attention. When we analyzed it, customer facing AI stuff gets 4 to 1 the attention of other things. When we look at the money invested, it’s the opposite 4 to 1 being funneled into risk and fraud applications. While you might be happy to say who your chatbot vendor is, you’re not happy to say who your cybersecurity vendor is because it probably helps hackers to know who you’re using.

Hurdles to AI adoption

The truest barrier to AI is a genuine understanding of what artificial intelligence is and what it does. It’s really hard to grasp these technologies. The only proper way to go about AI adoption is to think about what AI is capable of, what the realities and challenges are in deploying this stuff in the enterprise — you quickly realize how different and more complicated AI is from IT.

What are the representative use cases? What can AI even do? Unless the champion, the executive leader pushing this to happen is informed and has an AI transformation vision, they’re going to be disappointed, frustrated and ill-prepared for what’s to come.

When you go under the hood of an Ally Bank or Wells Fargo, their chatbots were dead eight months after being launched. They’re dead — you can’t even use them. That’s the lens of incentives at work. Without an informed yes, projects don’t have the support they need to succeed.