By John San Filippo
The 10th annual National Credit Union Collections Alliance conference was held April 2-4, 2024 at the Bellagio in Las Vegas. Finopotamus was onsite to learn more about how modern technology is being leveraged in collections and recovery. Perhaps surprising to some, the hot topic at the conference was artificial intelligence (AI).
AI Overview
In one session called Credit Unions and Fintech Partnerships: How Is It Going, Ashland, Ore.-based CU 2.0 Co-Founder and Chief Revenue Officer Chris Otey provided a broad overview of AI in banking and financial services. He noted that rudimentary AI and machine learning have been around for decades. “I think it was in 1998, I was working at Fiserv and we had an ATM controller and we successfully got it to recognize that, for example, every time ‘John’ went to the ATM, he took out a hundred dollars from his checking account, he didn't speak Spanish and he didn't want a receipt. The machine could learn that so that when John put in his PIN, his quick cash button would be that transaction.” He added that the earliest work in AI began nearly eight decades ago.
“So why is AI just now starting to take off?” he asked the audience, and then answered his question. “There are several factors involved. For one, the cost of computing has been going down for decades now. I've got more technology in my cell phone than I had in my college computer lab.”
To demonstrate his next point, Otey showed the audience a photo grid that consisted of chihuahuas and blueberry muffins (see below). “As recently as five years ago, a computer could not tell the difference between these two things,” he said to audience laughs. “So how did it get past that? Well, you have to feed the computer millions of images of chihuahuas and blueberries until it can actually figure out the difference between the two.”
Otey went on to tell the story of how he was playing cornhole at a party next to a gentleman who turned out to be the CEO of Pixar. “We got to talking about technology and how they're creating movies and how they're writing movies and how they're animating movies,” said Otey. “He said the technology has always been there for animating the movies, but Pixar is using artificial intelligence to write the movies. That was three years ago. The Pixar movies you see today are completely created by artificial intelligence. Sure, there are writers who doctor them up, but AI can write an entire movie, edit it, animate it, put the sound to it.”
AI in Fintech
Financial technology, of course, was central to Otey’s presentation. He showcased several fintech companies that leverage AI in different ways. “(Atlanta-based) Vertice AI has figured out that when we're marketing to members, we can use more data than we have on the core system,” he explained. “So if you're on CU*Answers or Symitar or DNA or Corelation or whatever core you're on, you have a certain subset of data, but you can augment that using third-party data and artificial intelligence to do a cash-flow analysis on every single member.”
Otey then provided a relatable example. When South Bay Credit Union, where he serves as board chair, needed to raise capital, the credit union decided to offer a money market account at 4.5%. “We were going to tell all our existing members, buy billboards, radio ads, Pandora spots, email everybody. The traditional way that we've always done it,” he said. “Vertice came in and said, ‘Don't do that. We've segmented your membership. We know which ones have a higher propensity to take that product. Let us target this 500-member group and we'll offer them 4%.’ Guess what. It worked because the artificial intelligence in the solution was able to combine data sources to make better decisions.”
Fraud is on the rise partly because fraud prevention and mitigation tools have not kept pace with fraudsters, Otey told the audience. “I sold a fraud prevention product at Fiserv (a company he left more than 10 years ago) and I guarantee there are still credit unions out there using it.” He explained that to counter the ever-changing tactics of modern fraudsters, San Francisco-based Effectiv uses “artificial general intelligence like a Tesla. Just as all Teslas ‘learn’ from any problem with one Tesla, every time fraud occurs at one credit union, every credit union in the country is alerted immediately. And because it's using the artificial intelligence, the changes are made on the fly to stop that fraud.”
According to Otey, competition among AI-centric fintechs is accelerating. For example, he noted that for several years, the only two companies in AI-assisted loan decisioning were Burbank, Calif.-based Zest AI and New York City-based Scienaptic AI. Now, he said, there are at least 12 companies competing in that space.
AI on the Front Lines
In a later session on AI for Collectors, Ashish Garg, CEO of Milpitas, Calif.-based Eltropy, explained how AI can be leveraged in collection environments. He began by making the case for the use of texting as a collection tool in lieu of email or physical mail. He told the audience about one Eltropy client, Canvas Credit Union, that is using texting effectively. “Here is the data that Canvas Credit Union collected,” said Garg. “When they sent a text message to collect – and let's say in that text message there's a link where a member can make a payment – they saw a 34% click rate on those links versus email or phone calls where they get half a percent click rate. So, texting in early stage collections is 68 times more effective than calling or emailing.”
Expanding on the idea of using texting for collections, Garg went told the audience that Eltropy is building ChatGPT-based AI system that will help collectors become more compliant, consistent and accurate in assisting members. “What we've essentially built is a framework of a large language model that takes in all the information from your credit union’s policies, procedures, documents, and even your websites, right,” explained Garg.
Garg and a colleague then acted out a scenario to demonstrate how this system might work in real life. A credit union collector texted a 45-day delinquent member requesting a callback. When the member called, he told the collector that his payment was late because he is ill. Using Eltropy’s AI system, the collector was able to instantly access the credit unions health-related loan repayment adjustment policy and generate a text message to the member that included a link to the required request form. Throughout the process, the AI system worked as a silent coach, ensuring that everything the collector said and did was in compliance with established credit union policies and procedures.
AI in the Back Office
In an interview after the session, Garg detailed for Finopotamus the full breadth of Eltropy’s new AI platform. “There are three different ways in which credit union can use generative AI technology,” he said. “One is for employees and that's what I showed on the main stage today. What I missed showing today was how you can use it for your members.” He explained that member-facing chatbots have historically be built using spreadsheets.”
“You take a spreadsheet with 700 questions, you add 700 answers and all this chatbot is doing is going through one by one to figure out what's the best answer to a member’s question,” he noted. “The problem with this approach is that spreadsheets are very laborious to maintain. The moment you upload one, it's already out of date because something has changed.”
In contrast, Eltropy’s new system maintains itself. “Our generative AI bot can crawl the credit union's website and other support documents by itself,” he continued. “So, the spreadsheet-based approach goes away. You just point the bot to your website and other member-facing documents.”
AI for CU Execs
Garg then turned his attention to the ways in which a credit union’s executive team can leverage generative AI. “Most credit unions take the pulse of what the members are asking for by doing surveys,” he told Finopotamus. “They do a biannual survey of the membership, but by the time the survey data gets collected, organized and shown to the executives, they're possibly nine months behind.”
The Eltropy system, however, can provide continuous feedback. “We have an ability through generative AI now that all these conversations that the credit union is having with the members – the phone calls they're having, all the text messages, all the emails, all the video meetings – we are able to record all of this and have generative AI parse through all these conversations and create daily summaries.”
Garg admitted that one problem with generative AI is that if it doesn’t know an answer, it will make up an answer. However, he added that Eltropy has built-in safeguards against this. “When it starts making up answers, it's what is called ‘hallucination.’ We don't want any hallucination.”
He noted that Eltropy’s generative AI platform is in various stages of pilot with 10 credit unions and will be generally available in early Q3, 2024.