Commercial lending team members at Great Southern Bank are using generative AI to “chat” with their data, especially policy documents and loan records. They ask questions like, “what are the loan-to-value limitations for multifamily real estate” or “what are the approval requirements for this loan” to get quick answers.
The use of the chatbot, nCino’s GenAI platform, is saving many hours of time and making the entire commercial lending department more efficient, according to Ryan Storey, director of loan operations.
Great Southern, which is based in Springfield, Missouri, and has $5.9 billion of assets and 89 branches across the Midwest, is one of many banks testing use cases for generative AI, a category of artificial intelligence that generates content based on patterns found in training data.
JPMorgan Chase has employees using OpenAI’s ChatGPT to draft emails and reports. Citizens Bank is creating an AI copilot to help call-center reps answer questions. Citi is rolling out GitHub Copilot to all its developers. Ally Financial is using generative AI to provide post-call summary recordings in contact centers. These banks are looking for efficiency gains and cost savings from their use of generative AI.
“In many cases, banks are choosing to do what I would describe as, ‘take the waste out and put value in,'” said Michael Abbott, Accenture’s global banking lead, in an interview earlier this year.
For instance, using gen AI to summarize customer calls saves about four minutes per call, he said. Mortgage loan originators use it to run loan applications through Fannie Mae requirements to find potential red flags. These kinds of operational uses can give banks 20% to 30% efficiency improvements, he said.
The use of generative AI to boost productivity in commercial lending is relatively new, according to Alenka Grealish, principal analyst at Celent.
“It’s a powerful use case and will likely gain traction,” she said. “Banks are continually on the quest to reduce process cycle time, which is a competitive differentiator. Hence, a generative AI tool that is relatively low risk — that is, it draws its response from internal documents — and expedites employee searches will be part of the current early adopter wave.”
Great Southern has been using nCino’s commercial lending software since 2018. This software analyzes loan applications and makes loan approval recommendations. Human loan officers review those recommendations and make the actual decisions of whether to grant loans.
Great Southern started rolling out nCino’s generative AI bot, Banking Advisor, in September. To train it, Storey’s team loaded loan documents into Banking Advisor. Then it fed the system more procedure-oriented documents such as training manuals and policy documents, including the bank’s 185-page credit policy. If someone wants to know what the loan-to-value limits are on real estate loans, they will type the question into Banking Advisor, which will return information without the user having to know which document holds the answer.
Using a large language model like this as a knowledge base does create another place where updated versions of documents need to be saved. Storey said this is not a burden as the bank has a process of ensuring there’s an accurate list of places where document revisions are stored. Great Southern first had its credit analysts test the system in a sandbox. Now the entire commercial lending team is using it, including loan officers and analysts, about 150 people all told.
One challenge that has come up is the variations in the way documents and people refer to certain terms.
“I might go in and ask what our risk rating is for a certain type of loan,” Storey said. The credit policy might use the term “risk grade” or “risk score.” Storey gets past this by training users to ask a question multiple ways, using different terminology, especially if they receive incomplete answers or no answer with the first attempt. The AI also references its sources so a user can quickly review the source material if the initial answers seem incomplete.
Storey has not seen Banking Advisor hallucinate — in other words exaggerate or manufacture answers — as many large language models do.
The fee the bank pays to use the tool is easy to justify based on the hours saved, Storey said. And he expects to reap more efficiencies going forward.
More recently, the bank has added another use case for Banking Advisor — generating narratives for commercial lending credit memos. The generator uses data from nCino and some predefined prompts to generate a narrative paragraph explaining the loan request, the collateral pledged or even a background of the borrower and related entities. This saves the credit analysts time and keystrokes writing narratives that contain information already input into nCino.
Accenture’s Abbott has seen other banks use generative AI to help create credit memos. He’s seen some provide generative AI copilots to commercial relationship managers, to help them analyze and recommend products like cash flow management and to help relationship managers hunt for new clients.
“It’s slowly but surely changing near every part of the commercial banking lifecycle,” Abbott said.
Storey’s team just launched an early look at another Banking Advisor tool that will review, sort and file loan documents into their appropriate locations within the electronic document manager in nCino.
“The chatbot knowledge base feature was really just the beginning of what we expect to accomplish with Banking Advisor,” Storey said. “We believe AI can play a role in enhancing how we conduct business.”