The MSU Financial Credit Union (MSUFCU) launched an artificial intelligence (AI)-powered external chatbot this year, but already the $6.6 billion credit union is planning to expand its chatbot automations in 2022.

MSUFCU continues to build its chatbots — Fran for customers and Gene for employees — but it will give them a boost with back-end robotic process automation (RPA) bots, Benjamin Maxim, vice president of digital strategy and innovation, told Bank Automation News.

“Identifying the processes that are maybe monotonous, that don’t need a person to actually do it — moving data from one system to the other — are usually pretty good candidates for RPA,” Maxim said. “We’re really to the point where we need to kind of prioritize which ones would make sense and which ones would maybe be quick wins.”

The credit union built out its chatbots using Boost.ai, a software company specializing in conversational AI. The Stavanger, Norway-based company’s client list includes $1.8 trillion Santander as well as $33 billion DNB and $66.9 billion Sparebank 1 SR-Bank, both Norwegian banks.

Boost.ai offers a library of pre-package intents to go with its industry-specific models, Bill Schwaab, vice president of North American sales, told BAN.

“The beauty of having these industry models pre-built is that when we do get to the point where we want to add, say, money movement or other authenticated user-type activities, those models are built, and it’s essentially just configuring them to align with Michigan State’s specific needs, and the tone that they want to have with their customers,” Schwaab said.

Gene, an internal chatbot designated for employees, launched in early 2020 with a different vendor. Fran, an external-facing bot for customers, came online in early October with Boost.ai, Maxim said. Fran helps support the credit union’s chat function, which grew from a department of two into 85 employees over 14 years, Maxim said.

The credit union also faced a growing challenge of customers moving to other time zones after graduating from Michigan State University, he said.

“Our call center closes at 9 p.m. eastern. If you’re getting off work at six o’clock in California … you’re probably not able to get the help you need,” Maxim said. “So Fran was an opportunity to expand to 24/7, 365 [day] support.”

Before Gene and Fran, help-desk employees searched SharePoint documents to find the proper response to questions. That would mean long hold times for customers, Maxim said.

“If you don’t have the right keywords, or if the documents are flagged the right way, it’s hard to get a response, and then you have to read through perhaps a 30-page document,” Maxim said. “We averaged about eight minutes of time that was spent in those interactions where the member was on hold while the employees were going through the answer.”

It was also a challenge for the bank’s e-services team, which handles the live chat rather than calls, except it averaged a five-minute wait for a response via the chat room, he added. After a four-week pilot with Boost.ai, the credit union realized that it could automate about 2,000 interactions in its chat room.

MSUFCU started by reviewing its topic documents, and over 10 weeks built out the rest of the knowledge base, Maxim said.

“The greatest thing about chatbots is people ask what they think it should answer,” he said. That allows the team to evolve the chatbot by reviewing conversations and adding missing content.

Boost.ai also helped train individuals who had no prior AI experience, Maxim said. Maintaining the bots was a part-time job until the credit union established a new position to support its chatbots — AI content manager.

“You need to update the AI,” Maxim said. “This content person is able to review the conversations that are had with Gene and Fran, see what was answered correctly, what was not, what can be improved, what training needs to be done to improve the data behind the answers to improve the overall satisfaction moving forward.”