By John San Filippo
Fintech startup Stakana is on a mission to bring world-class, artificial intelligence (AI)-driven predictive analytics to credit unions. Finopotamus spoke with Nate Derby, the company’s co-founder and CEO, to probe Stakana’s value proposition in greater detail.
Roots in Retail
“My co-founder, Charles Brophy and I had worked in the retail industry for about 20 years each,” Derby told Finopotamus. “He wrote the recommendation engine for Zulily. I did customer loyalty for Nordstrom and customer attrition forecasting for T-Mobile. We also both worked for different marketing agencies. But when we both hit 40 [years old], we decided we wanted to do something with our lives a little more impactful than making more money for Amazon or Starbucks.”
Derby and Brophy had both been credit union members their adult lives, so they understand the positive impact of credit unions. “I'm well aware that credit unions, especially the two thirds of them that are smaller, like below $100 million in assets, are really, really struggling,” said Derby. “The larger ones are not, but we want to be help them all by bringing this technology to the table, really demonstrate an ROI and have an impact.”
Struggling With Data
According to Derby, credit unions have long been interested in leveraging data, but there are many ways these efforts can fall short. “You can hire the right people and have them managed badly,” explained Derby. “You can work on the wrong projects. You can work on the right projects, but not using the right methods. I met one credit union that hired two Ph.D. people and didn't know what to do with them. Another one got a great report and then quietly shelved it. They didn't operationalize it.”
This last point, he added, is critical. For data to be of any value, credit unions must operationalize any insights that are derived from that data. “We are focusing on the predictive part because that's the hardest part,” said Derby. “So many credit unions can hire the right people to do data warehousing, to do cloud computing, to slice and dice the data, make a data lake, come up with reporting. But the real value that we think we can push out there is the predictive part.”
Predictive Analytics in 2023
While predictive analytics have been employed in business for decades, Finopotamus asked Derby what makes AI-driven predictive analytics better than previous models.
“Ten or 15 years ago, we used standard statistical techniques that have been around for 50 years – methods like linear regression or time-series analysis,” explained Derby. “These can work, but the underlying assumptions are not that flexible. The underlying assumption is that behavior is governed by a process that either doesn't change that much or changes just a little bit over time. But when something happens that changes behavior quickly, the statistical models are relatively slow to adapt, and that makes for the bad predictions.”
Derby noted that COVID-19 offered a perfect example. “The pandemic was a huge shock. And for a couple months, every predictive engine out there under the sun was predicting badly. Then two to three months later, it all converged and everything was back on track. That’s the real difference with machine learning.”
He added this type of technology takes a lot of computing power, but because of the rise of low-cost cloud computing, the resources are now available to make the technology affordable for even smaller credit unions.
Uses Cases
What might a credit union want to predict? “The first thing they might want to predict seems kind of basic,” said Derby. “When is a member thinking of closing an account? You can always tell when somebody turns off direct deposit. That means the member is already out the door, so maybe we'll just call them and try to win them back. But by that point along the member journey, the member's already out the door. Maybe he or she hasn't closed the account yet, but the decision is made.”
Derby claimed that Stakana’s system can predict that potential exit much earlier. “Our objective is to identify earlier in the member journey when a member is thinking, starting to change behavior, because it's much easier to correct at that point,” he said. “The member might not even be thinking about leaving consciously. And if you can do a few very subtle corrective actions earlier in the journey, it's much more effective.”
Stakana conducted a study with Peoria, Ill.-based Citizens Equity First Credit Union (CEFCU) ($7.5 billion, 388,000 members) to validate the effectiveness of this approach. Derby described the study methods:
“Every week we forecasted, these are the members that we think are going to leave in three to six months. Next what we do is we randomly select a few to be in a control group,” he said. “Then three to six months later, we go back to see, of those that we predicted would leave, which ones actually left versus the ones we didn't predict.”
There are two different measures of success: Was the prediction accurate and was any invention effective? In terms of interventions, Derby said it’s important not to be “creepy.” He noted that CEFCU created a subtle intervention for such cases, which he described as “surprisingly effective.” He said, “What they did was send the member something by physical mail asking them to fill out a survey. Turns out hardly anybody filled out the survey, but the people who got the survey statistically have been shown to have stayed on.”
Developing Indirect Members
Stakana is currently developing a new product that uses its predictive analytics to help cultivate indirect members into more engaged, full-fledged members. “Credit unions have been trying since the beginning of time to make indirect members into standard members. That generally hasn't worked,” stated Derby. “But what if you can find which indirect members might be the best candidates for the next level up for a financial product?”
The issue with indirect members is that the credit union doesn’t have the trove of transactional data that it has for other fully engaged members, he explained. Stakana, according to Derby, has solved this issue. “You actually use some clustering techniques to look at the attributes of the indirect member and compare it to people with similar attributes who are regular members and see if you can make a match. It's a little bit out there, but it is an approach that is doable.”
Stakana, Derby noted, is currently seeking credit unions to pilot this program.