AI Report Digital - Flipbook - Page 11
Serving Customers
In 2021, East Lansing, Michigan-based Michigan State
Financial institutions are increasingly using AI to assist cus-
University Federal Credit Union launched a virtual customer
tomers. Bank of America Corp. was an early mover; its AI
service agent named Fran in partnership with boost.ai. Fran
assistant, Erica, handled 170 million interactions in the third
allows the credit union to provide around-the-clock service,
quarter of 2023.
resolve common questions, and help human agents manage
workflow and focus on more complicated inquiries. Within
two weeks, the chatbot resolved 81.1% of customer inquiries successfully; that improved to 98% by October 2023
with assistance from the credit union’s AI trainer team,
according to the Norwegian software provider.
And before it was acquired in 2022, Sterling National
Bank, the bank unit of Pearl River, New York-based Sterling
Bancorp, used an AI assistant in its call center to authenticate customers and manage simple inquiries, like sharing
account balances and recent transactions, troubleshooting
online banking issues and resetting passwords, handling debit
Fran is the first point of contact for members who use the
card claims, managing declined card transactions, confirming
credit union’s live chat service; it handles the equivalent chat
fraudulent charges and providing nearby branch locations.
load of nearly 60 full-time employees across almost 45,000
Customers could still interact with a human. Over time, the
conversations a month. Application programming interface
software learned from interactions so it could better manage
(API) connections automate 10 common account processes
calls without human agents. That had Sterling’s AI assistant
so Fran can share personalized information on rates, account
managing about 100,000 calls a month.
balances, payments and loan payoffs.
Sandboxing AI
In 2023, Winter Haven, Florida-based SouthState Corp.’s
“Tate essentially unlocks valuable information stored within
internal innovation group, dubbed Spark, created and
our bank and makes it accessible to all,” Nichols wrote.
launched its own chatbot to compete with its internal
“Training time gets reduced, productivity increases, and
intranet. Comprising 35 employees from across the bank,
employees provide more accurate and helpful answers.” He
Spark brainstormed this use case for generative AI technol-
also pointed out that seven minutes can feel like “an eterni-
ogy, outlined what it would be like for someone to use the
ty” if a banker is with a client.
technology and developed an action plan, according to an
article authored by Chris Nichols, director of capital markets at the $44.9 billion bank.
Tate’s answers include the model’s thought process, supporting content and citations for each answer, which allowed
bank testers to rate the AI assistant’s accuracy for 1,600
The result — a generative AI chatbot called Tate —
questions. That helped developers track the model’s process
functions as an internal librarian for employee queries.
and increase its relevancy.
Employees can ask Tate questions, which it answers using
approved SouthState documents, rather than needing to
search for the right documents and policies on the company’s
intranet and then searching those documents for the answers.
Tate reduced the time it took for employees to locate information from seven minutes to about 32 seconds, according
to Nichols. Correctly phrased prompts produced answers in
less than 15 seconds.
SouthState built the chatbot in its Microsoft Azure environment, using only internal data. It cost the bank about
$25,000 to build and about $50 a day for 100 employees to
operate. At that rate, 100 users asking Tate two questions a
day would cost the bank about $13,000 annually but save
about 5,200 hours and up to $442,000 of employee productivity, Nichols calculated.
ARTIFICIAL INTELLIGENCE: A REAL-WORLD APPROACH | 9
