Banking technology provider CSI has developed a three-tiered AI deployment framework designed to maximize productivity gains while managing runaway token costs. 

The Kingsport, Tenn.-based company is using AI to: 

• Customer-facing tasks; 

  

• Boost operational efficiency; and  

• Improve internal productivity.  

CSI also is being strategic about which LLM to assign for each AI application to manage token cost, Michel Jacobs, chief strategy officer at CSI, told FinAi News. 

For high-complexity tasks including product requirement documentation, dependency mapping, legacy application analysis and advanced code development, CSI uses Anthropic‘s Claude, Jacobs said. 

For development tasks including UI construction, API specification writing and MCP development, the company leans on open-source models such as Meta’s Llama, Jacobs said. 

And Microsoft‘s suite of AI tools handles general office productivity across its roughly 1,600-person workforce, he said. 

“If everybody will use full Claude for the basic stuff … your token bill will go up,” Jacobs said, adding that the company has assigned tasks to models that play best to their strengths. 

SUBHEAD 

CSI’s AI tiering is task-driven rather than seniority-driven, he said. 

“We have junior developers that are 100% more efficient than senior developers” because of AI use, Jacobs noted, pushing back on any assumption that frontier model access is rationed by role. 

Jacobs said CSI has not yet reached a threshold where its chief finance officer is sounding alarms but acknowledged that the cost trajectory in the industry is what prompted the structured approach in the first place. 

FIs that use CSI’s solutions include: 

• Home Federal Savings Bank; 

• Seacoast Banking Corp. of Florida; 

• First Keystone Financial; and 

• Citizens Community Bancorp. 

QuickBooks machine learning integration 

CSI is seeing rising developer efficiency of as high as 30%, Jacobs said, adding that AI is helping to more quickly develop drafts and prototypes. 

At the same time, CSI is building an automated payment reconciliation layer that would connect directly into small business accounting platforms including QuickBooks, Xero and NetSuite. Jacobs said, with a release targeted for this year. 

The integration would use a combination of API calls and machine learning to match incoming ACH payments against open invoices in real time — automatically posting entries without manual intervention from business owners, Jacobs said. 

He added that AI can kickstart a new era for commercial banking, which has lagged in digital features compared to retail banking. 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.