Previous Everyone says they’re using AI. Few are trusting it with decisions that matter, like borrower approval. Allison LaForgia talks with First Colony Mortgage’s Corey Shelley and Tidalwave’s Diane Yu about AI deployment, where the stakes are real: 700-page loan files, compliance pressure, and zero room for error. Corey Shelley and Diane Yu break down how they built AI to handle one of the most complex parts of lending, and why it’s the complex part that counts.
Shelley described the initial challenge that led to the collaboration. “With wholesale, we were growing so fast, and margins are thin and scalability is hard,” he said. “What we saw with Diane’s product is it’s a way that we could scale, cut out a lot of time . . . without having just to keep hiring and hiring more people.”
Yu explained that the partnership was intentionally built around real production, not theory. “I never really like building the solution in the ivory tower,” she said. “We worked together for a year . . . looking at the production volume, how to build the right solution for wholesale.”
From there, collaboration quickly scaled in scope and volume. “Now 100% of wholesale volume from Corey’s wholesale division is going through our Tidalwave solution,” Yu noted.
For Shelley, that translated directly into immediate and measurable business impact. “We went from doing very little wholesale volume when we started, all the way up to 600 million within about a year and a half,” he said. “The scalability it creates and the time saving . . . is like rocket fuel for our company.”
At the center of that impact is a fundamental shift in how loan files are handled. “Wholesale loan packages are anywhere from 500 to 1000 pages,” Shelley said. “We call it ‘stare and compare,’ organizing them out . . . Tidalwave took those 1000 pages, immediately organizes them for us.”
But the transformation goes deeper than document organization. “It starts to correlate all the different [data] . . . the credit, the income, the bank statements,” he explained. “You kind of have to break how you think about mortgages . . .underwriting starts the second the borrower uploads everything.”
With underwriting starting as soon as documents are uploaded, early feedback is immediate. “The response from teams has been strong. ‘My loan officers love it. I’m getting messages and texts constantly saying, “Hey, this is the greatest thing that we’ve ever used,” Shelley shared. “My underwriter is saying we’re saving so much time . . . the AI is showing us everything we need to see immediately.
He added, “So those are your two probably highest-cost centers, and you’ve just reduced their time to be more efficient and work on other things.”
Yu emphasized that this level of performance only comes from real-world iteration. “There’s a huge difference between a demo and the real production product,” she said. “A production product . . . [must] take all those complex scenarios to be able to be laser accurate.”
She added that compliance and accuracy are foundational to Tidalwave’s success. “We’re building a capability on top to make sure all the questions that we answer are in compliance, and all the information we provide to the loan officer is accurate,” Yu said. “It takes a team . . . to truly build a system to make this work.”
On the ground, the day-to-day benefits for every part of the loan process are clear. “It’s the ultimate time saver . . . from the loan officer all the way through closing,” Shelley said. “We’re finding the problems way early . . . instead of at the very end.”
Looking ahead, Shelley sees even more opportunity. “I feel like we’re just scratching the surface on what’s possible,” he said. “Watching the team think differently and finding better ways to do mortgage because of the Tidalwave product.”
Yu added that the wholesale use case expands what’s possible for the broader industry. “You’re going to see that our agent suite can be deployed as a standalone solution,” she said, enabling lenders to support processors and underwriters at scale.
Together, their partnership points to a shift away from traditional mortgage workflows toward a faster, more integrated model built on real production AI.
