When customers contact Wells Fargo, AI system goes to work

Register now

While many banks are stepping up their use of chatbots and virtual assistants to keep in touch with consumers during the pandemic, Wells Fargo has taken a different route.

It has built a system it calls Advanced Listening that lets customers communicate through the channel of their choice — phone call, email, text, survey response, online banking interaction — and listens to or analyzes every interaction. The system uses artificial intelligence to understand what customers are saying, identify problems or needs, root out widespread issues and provide help more efficiently.

This lets the bank do several things. It's starting to automate responses to high-volume, low-risk complaints. It alerts lines of business if it’s getting complaints around a certain theme. It detects potential compliance violations or systemic issues. It can guide customer service representatives to respond intelligently to customer queries. It can also be used to create useful applications, such as customer journey maps for loan officers. A customer journey map is a diagram that illustrates the steps a customer goes through to accomplish something, such as obtaining a loan.

Wells Fargo is a company that has been trying to repair its relationship with customers. For about 14 years, starting in 2002, employees in the community bank division were pressured through aggressive sales goals into opening millions of unauthorized accounts and issuing millions of unauthorized cards, according to the Office of the Comptroller of the Currency.

The bogus accounts were uncovered by the Los Angeles Times in 2013 after it received dozens of complaints from Wells Fargo customers. The aftermath included government probes, $185 million in penalties in 2016 and a federal cap on assets that is still in place. Some customers must have complained to the bank first. If it had the ability to feed all customer communications into one AI engine, perhaps it might have caught and addressed the problem.

Wells Fargo doesn’t position Advanced Listening as a reaction to the scandal.

Michael Soistman, senior vice president of enterprise complaints data analytics and reporting at Wells Fargo, describes Advanced Listening as an alternative to banks’ traditional virtual assistants.

Michael Soistman, senior vice president, enterprise complaints data analytics and reporting at Wells Fargo
"Customers’ interactions with banks can be complex and even emotional," says Michael Soistman, senior vice president, enterprise complaints data analytics and reporting at Wells Fargo, whose team built the bank's new Advanced Listening system.

“Those are very interesting, great technology,” Soistman said, noting that they also put a layer of technology between the customer and the company. Customers don't always want to speak with a machine, especially when it comes to complex and emotional matters, he said.

"Your machine may prompt you for questions that are not necessarily aligning to how the customer is thinking or talking," he said. "We've taken a different approach to a very similar interaction. We've preserved the customer’s decision to communicate in whatever approach they want to use. They can use a digital channel if they want, or they can talk to a customer service rep. There's no incremental effort or impact to the customer.”

When pressed, Soistman acknowledged that compliance was a factor in the creation of Advanced Listening.

“This capability was not launched as a result of a regulatory requirement, but we as a company are very focused on addressing our risks and controls and continually improving customer experience,” he said. In banking, “regulatory activity can aid improvements by focusing attention and investment, and you can really build additional momentum on related innovative ideas. This program certainly benefited from that since it supports these ideals.”

What Advanced Listening is

Advanced Listening has three layers, Soistman said: data, analytics as well as business integration and insights.

As conversations take place, Advanced Listening monitors both the conversation and what the team member does during that conversation. It tries to understand and categorize the gist of every interaction. Then algorithmic models look for changes and emerging trends.

“Every day we're monitoring the communications we get — this adds up to millions of communications each month,” Soistman said.

In a recent example, Advanced Listening uncovered a problem with a third-party vendor that a business customer was using to deposit payroll funds into Wells Fargo bank accounts.

The intermediary company had a breakdown. Meanwhile, employees expecting to get paid called their employer and Wells Fargo to complain. The Advanced Listening system identified the spike in complaints, which led to management paying attention to the problem, quickly researching it and figuring out a resolution.

Today, this does not yet happen in real time, Soistman acknowledged.

Soistman said he hopes that in the future customer service associates will be told in real time what is happening and the best way to respond.

“That's an aspirational goal,” he said. “What we can do today, though, is find emerging trends and quickly notify the business. We're able to categorize and analyze the interactions and even automate aspects of our complaint management process.”

Another use case for Advanced Listening is to listen to phone calls to make sure certain compliance activities are performed during the phone call.

“The vision throughout has been to create a capability that could improve compliance, controls and customer experience while reducing associated costs,” Soistman said. “I think this capability can ultimately help transform the way we service our customers, through bringing real-time insights to the front line so they can engage with the customers in a more prepared and organized way.”

Soistman said it was necessary to build the system in-house so that it could be easily pivoted for different use cases. The bank also felt it would be quicker to build internally.

Soistman’s team, which includes data scientists, data architects, project managers and business process experts, relied heavily on open source algorithms to obtain the patterns and insights. They used the Python programming language.

Agile methods were used for much of the development, which took a year and a half. The system was turned on this year.

Data challenge

The data layer was the hardest part of building Advanced Listening, Soistman said.

“Aggregating and preparing that massive amount of data was the most difficult challenge and certainly required the largest investment of resources,” he said.

Phone conversations, for instance, have to be recorded, transcribed and fed into the system. Those are matched with any interactive voice response actions the customer made during the call, which helps categorize each call. They’re also combined with any actions the employee took during the call, such as opening an account or charging a fee.

“When you bring all that information together, you develop a holistic view into what happened in the conversation,” Soistman said. “That's our utopia. The challenge is pulling all the data together, figuring out how to apply it so you can glean insights from it. Every channel is different, and each line of business can have differences that need to be considered.”

The outcomes

If the system uncovers a theme, such as a rise in complaints about remote deposit capture, that will be advanced to the line of business for investigation and resolution.

“If there's something that we identify as a potential systemic issue, we have a process in place to forward it to a research team to study whether we have a systemic issue on our hands that needs to be addressed,” Soistman said.

The bank has begun automating the handling of simple complaints. It has also started building applications that work with Advanced Listening. For instance, one potential application would create customer journey maps for loan officers so they can see what’s going on with their customers' loan applications.

“If a customer is going through a loan application process — providing documents, answering questions and engaging in the related back-and-forth review — we are able to track communications and other activity through their journey, and we're able to score it based on this information,” Soistman said. “We think that could help loan officers prioritize which customers they reach out to."

That could help the loan officers prioritize which customers they reach out to.

“We've been experimenting with some lines of business on visualizing this concept through dashboards that help the business easily prioritize,” Soistman said. “The idea is that managers can use this to prioritize their engagement or otherwise intervene when customers may be experiencing a high level of frustration or other operational challenges.”

Soistman said he expects many more uses cases to come out of the woodwork.

For reprint and licensing requests for this article, click here.
Customer experience Customer service Artificial intelligence Wells Fargo Compliance Digital banking Digital Banking 2020