Wall Street is racing to deploy AI across underwriting, risk, and wealth management, but banks are running into an operational roadblock: capturing decades of unwritten expertise held by senior bankers and turning it into documented playbooks. On Market Movers: Midday, Remy Blaire speaks with Leo Rocco, founder and CEO of Confidence Systems, about why extracting what sits in employees’ heads is the real choke point in enterprise AI, especially as regulators demand to know how algorithmic decisions are made.
Rocco says much of the existing documentation is too assumptive to support building agents. He explains that Confidence Systems forward deploys engineers to capture how people actually perform work down to the clicks, including the happy path, its variations, and the exceptions where he says banks and financial institutions do much of their business. He says the company can capture that knowledge in 30 minutes or less and produce process maps and playbooks. Rocco says many banks are spending heavily out of fear of not acting, with value that is hard to measure, and that auditability is a challenge because there is often no good audit trail and not enough humans in the loop.
Rocco argues that companies should own their data, capture it locally, and build agents on top of it without paying token fees to large LLM providers. For institutions deploying generative workflows, he recommends a crawl, walk, run approach: understand how the organization operates, have humans validate workflows, introduce agents one task at a time with humans in the loop, and make sure there is meaningful return. He adds that in many cases, having humans do the work is just fine.