Scarpero says taking advantage of the latest AI technology could free up hours for other LOs
Artificial intelligence has moved past drafting social posts and answering borrower questions for some mortgage professionals. Work that back-office staff handled that didn’t require a license is increasingly being automated.
Increasing automation does come with some concerns. Turning a computer program loose on a CRM full of borrower data, or on outbound emails to real estate agents, comes with compliance and data security risks that must be managed through both software guardrails and human oversight.
One broker, who has been on the front lines of using new AI technology to streamline his business and increase his outreach, has spent the past several weeks testing new tech. He is treating the software less like a chatbot and more like a new hire he has to train and supervise.
Carlos Scarpero (pictured top), a mortgage broker at Edge Home Finance, has been running an AI platform called Viktor through some of the more tedious parts of his business. That includes cleaning up years of mislabeled borrower records and drafting a detailed dispute letter to a real estate agent.
"Viktor is basically like having a true assistant, but it's a computer," Scarpero told Mortgage Professional America. "Basically my job is the editor and monitor."
Cleaning up years of data
Scarpero said one of his first real tests came from his own CRM system, which was not keeping pace with his loan origination software, ARIVE. Borrowers were landing in the wrong pipeline stage and getting messaging that did not match their actual file status.
His CRM used a single status called "adverse" to cover both borrowers who had withdrawn and borrowers who had been declined.
"How do I differentiate that automatically without me doing a lot of manual labor?" he said. "Nothing worse than someone that withdrew the file or got declined due to their 75% DTI but has a 700 credit score. That's a different set of messaging than the person that has bad credit."
Scarpero said he had Viktor audit his CRM. The tool found roughly 25 files sitting in the wrong stage.
Viktor eventually corrected about 85 contacts total, splitting them into separate credit repair and general decline pipelines.
While Viktor’s increased capabilities over simple AI chatbots have been beneficial, it also requires Scarpero to keep a close eye on it to make sure it’s following mortgage regulations in what it can and cannot do.
He said he sets hard limits on what the tool is allowed to see, and the tool has built-in guardrails as well.
"Don't give it a Social Security number. Don't give it ARIVE data," he said. "Don't give it your ARIVE password. Don't be stupid."
Those limits were tested. Scarpero said the tool once asked whether it should pull data tied to a borrower's credit report. He pushed back on the request. The tool then told him it could not actually access that information.
"I don't want it to access that because that would be illegal if it could and that'd be a big problem," he said.
Not AI slop
The same tool, Scarpero said, has taken on far larger jobs. After Edge Home Finance sent him a 15-page corporate compliance checklist ahead of an internal audit, he uploaded it to Viktor along with a single instruction.
Viktor reviewed roughly 800 pages across his website and social accounts, identified dozens of compliance and SEO issues, and corrected them after Scarpero told it to fix what it found.
It also rebuilt nine calculators he had been paying a third-party vendor to host on his site. That let him cancel the subscription.
A real estate agent sent Scarpero a contract asking for concessions he believed violated lending rules. He had Viktor pull the property's appraised value from his email history, compile an exact repair list and draft a response citing the relevant VA handbook sections.
"I don't want to send some blind thing to her and not like it," he said. "I read through it. I'm like, ‘Oh man, this is really good.’ Do you know how much time that just saved me and hassle that just saved me?"
Scarpero pushed back on the idea that AI-generated work amounts to what critics call AI slop, arguing the label misses how he actually uses it.
"You’ve got to treat this like you treat a real assistant," he said. "You're not going to tell your assistant, ‘Hey, go create blog posts for me.’ I mean, you could, but that'd be stupid. If they're not in the industry, they're going to create a bunch of garbage. So why are we treating the AI any differently? I still review it and approve it."
Scarpero estimated he is paying about $4,000 a month for the platform. While that may be more expensive than other platforms, he estimates it would cost more to hire people to do the same job.
"If I didn't have this technology, what would be the option? Going and hiring someone or multiple people," he said. "That would cost way more than $4,000. Probably $10,000 to $15,000 at least to do it at the level I'm doing it. So from that perspective, that's actually a pretty good deal."
Stay updated with the freshest mortgage news. Get exclusive interviews, breaking news, and industry events in your inbox, and always be the first to know by subscribing to our FREE daily newsletter.
