Valuation, Optimization & Attrition

How machine learning and AI enhance human interactions.

By Alaina Webster, Managing Editor

If searching for a term to describe the recent financial landscape, the Great Data Reckoning may be apt. By now, we all know how much “data” banks have — there have been countless articles written about untapped data, unanalyzed data, data not meeting its potential. If data were a sullen high-school senior with good ACT scores but mediocre grades, it would be sitting in the guidance counselor’s office right now, being asked why it can’t see what a great contribution it could make to the world and whether it would consider adding a few extra-curriculars to its resume.

Likewise, we’ve all read that machine learning and artificial intelligence are here to help data, assisting it in being all that it can be in the world of finance.

Marvin “Mickey” Goldwasser, vice president of marketing at Glastonbury, Conn.-based Payrailz, sees the combination of data, machine learning and AI as “almost like hitting a reset button … We get to start over with a blank slate, with the following question: What would you do differently with the tools that are available today, with the knowledge that you have today?”

For Robert Stillwell, business analytics officer for Seacoast Bank, headquartered in Stuart, Fla., “Everything is about generating value between the bank and the customer,” meaning that analyzing data holistically at the customer-level is more strategic than examining account-level or transactional information. “What’s the value of every single customer?” he said. “How valuable are they, and why?”

In order to find out, Stillwell and Seacoast “wrangled” the data, creating “about 27,000 columns of data for every single customer,” Stillwell revealed. Seacoast then partnered with Cary, N.C.-based SAS to institute first software and then a server-based environment that allowed the bank to move from data gathering to interpretation to predictive analysis using machine learning algorithms in the span of approximately five years.

The bank has developed three key uses for the data fed through SAS’s platform: an opportunity sizing engine, branch network optimization and predictive value loss.

Because Seacoast’s bankers can see the “value” of a customer as a whole, they can identify gaps in the banking relationship. For example, Stillwell said, a dentist who has a very healthy deposit record and a loan with the bank is a customer generating obvious value — however, the customer is not efficient to serve because she makes a large number of in-person deposits, requiring more time from tellers than the average customer.

“You don’t want to tell the customer to stop visiting the branches,” Stillwell said, “but you might want to let them know about the convenience of a remote deposit capture machine … What the opportunity sizing engine does for every customer is help our marketers and sales folks zero in on where the value is for that customer. Where’s the next level of value coming from?”

Stillwell described the bank’s need for branch optimization thusly: “In the industry, there’s been this digital transformation. Everyone is scaling back their branches, and we’re using machine learning to do this in a very intelligent way. We’ve built machine learning models based on branches that we’ve closed in the past. We have insights as a result of those closures.”

Feeding this data into the SAS platform, analysts can determine the impact of closing a specific branch. The algorithm factors in how many other branches are in the vicinity and combines customer scores and branch scores to ascertain how closing that a banking office would affect customer lifetime value. Would customers be able to access another Seacoast branch at a nearby location, or would the bank risk customer attrition because the closure would make it inconvenient for locals to continue using the bank?

Human input is always still a factor, Stillwell acknowledged. “The branch closure analysis required judgement to determine and calculate the variables to feed into the model. The number of nearby branches, for example, wasn’t something that existed natively in data sources. I had to go out and get the addresses of all the branches, and that was painstaking, but we intuitively felt it would be a good predictor.

“The algorithm ultimately evaluates whether a variable is a good predictor or not,” he continued, but there is a judgement. “What could be good predictors, and what could I feed into the algorithm? That’s where I think the value of data science is going forward, complementing technical knowledge with intuition and judgement.”

As for predictive value loss, Stillwell called it “a little bit of an epiphany.” Traditional customer attrition models focus on low-value customers because they are likely to attrite, which Stillwell described as “trying to save a relationship that wasn’t really a relationship to begin with.” Seacoast’s adapted predictive value loss model looks for customers most likely to have a significant and sustained drop in value, rather than only those likely to leave.

As a treatment against customer and value loss, the bank is using the evidence gathered to proactively contact customers and ask if there’s anything they’re missing in the banking relationship. It’s not a sales call, Stillwell pointed out. The call is intended to be purely customer centric.

“In the industry, we could call that next best communication versus next best action,” said David Wallace global financial services marketing manager for SAS. “It could be financial advice, for example. If you know all the transactions, you could reach out to, let’s say, a millennial and say ‘You’re halfway through the month, and we see that your rent is coming due in 15 days. We also see your entertainment spending is 30 percent higher than the last three months’ average, so you might want to watch that.’ That’s not a Big Brother thing; that’s simply flagging something that the customer could determine independently but likely isn’t focused on.”

Data presentation can be scaled according to who’s using it, their level of technical knowledge and their position within the banking organization as well.

“We call it approachable analytics,” said Wallace. “You’re showing visualizations of the data and the customer lifetime value process, so the bankers or marketers can determine what actions they need to take — but it’s not presented in technical terms. It could be a chart or a graph, or for a forecast, a simple line going up or down. Even a nontechnical senior banker can understand that forecasts should go up, not down.”

Wallace continued, “SAS Visual Analytics doesn’t require any knowledge of programming. For example, a business analyst at a bank … might not know the details of how to apply the algorithm, but they understand relationships between the data. That analyst could simply grab some data, drag it across to a pallet and click some icons to get the forecast for that data. That’s what we most often see in banks that use the software. People like Rob [Stillwell] do the data wrangling, and they deliver data their personnel can interact with. It’s more of a guided approach.”

Data has become even more central to customer engagement as banking interactions increasingly move to digital platforms. “If you never see the person, then you have to let the data and the analytics identify what is that thing that the person is needing,” said Wallace. “We would call that real-time customer experience. When the marketing people are building all of that out, they need to track the customer journey. What are the segments that are doing which customer journey where they navigate the websites and mobile banking? You can use machine learning to do what’s called automatically derived segmentation.”

And customers are coming to expect certain interactions based on the data financial institutions possess. “The thing with banks and credit unions is that with your permission [emphasis his] they’re going to use your data,” said Goldwasser. As an example, he offers Amazon’s Alexa assistant. “Someone says, ‘Oh my god, Alexa’s listening to you!’ Well, I hope so because when I say ‘Turn the lights on’ I want the lights to go on. It’s not technology for technology’s sake, but technology because it solves a problem … The issue is we’ve got all this data, what are the good applications for it? Why not use the data to make servicing your customer or member more engaging and predictive?”

As always in finance, however, those two watchwords — regulation and compliance — must be considered. Far from being the “black box” of myths and legends, AI and machine learning can be made transparent, depending on which system you choose.

“If your bank wants to do something that needs regulatory compliance, regulators will require that you explain how the model did what it did,” explained Wallace. “With less transparent models, the data goes in and the answer comes out, but you can’t describe the process. That’s not going to pass muster.”

At times, he said, “The decision is not to pick the best thing; it’s picking the thing that’s most likely to get regulatory approval.”

Despite compliance and regulation challenges, Stillwell and Seacoast see tremendous value in the system they’ve created.

“Maybe 10 years ago, all the data was accountables,” Stillwell said, “The deposit is served by this guy, and the loan is served by that guy, and I think a lot of banks are still struggling with that … Now we see things holistically.

“There’s a lot of judgement and also collaboration with the business to understand what decisions are currently being made. How are they being made? What challenges are folks having when they’re meeting face-to-face with their customers? The more you have that context, the better you can do.”

Alaina Webster, Managing Editor,

  • Sign Up

  • Categories

  • Archive

Software: Kryptronic eCommerce, Copyright 1999-2020 Kryptronic, Inc. Exec Time: 0.060125 Seconds Memory Usage: 3.807922 Megabytes