Pennymac is rolling out Google Gemini enterprise-wide to secure internal data while boosting productivity, according to National Mortgage News. Senior Managing Director and Chief Information Officer Mike Hogan said that the initiative allows staff to utilize artificial intelligence safely without exposing sensitive consumer or corporate records to external training models.
How Pennymac deployed Google Gemini
Pennymac adopted Google Gemini enterprise-wide to close specific workflow gaps where employees attempted to use unauthorized external artificial intelligence applications. The lender established a secure, isolated data environment for corporate tasks to prevent sensitive information leaks.
“We were trying to make sure people had access to tools to help them write better spreadsheets, documents, emails, and people had started saying, ‘Oh, can I bring my own ChatGPT to work?'” Mike Hogan, senior managing director and chief information officer at Pennymac, told National Mortgage News.
Hogan said that initial internal hesitation slowed early adoption among workers handling sensitive files. Management addressed this by removing restrictions on internal tasks. “That was, again, a real unlock for us to just tell people use it completely,” Hogan told National Mortgage News.
Why thorough training drove adoption
Pennymac implemented extensive staff training programs to move beyond basic tool distribution. Hogan explained that relying solely on self-motivated employees results in limited usage across large organizations. “If you really want full adoption, you really need them to train them,” Hogan told National Mortgage News.
The company provided peer examples and modification frameworks to help personnel integrate the technology into daily operations. This structured instruction helped staff overcome initial learning curves and extract practical utility from the software.
What testing flaws delayed the rollout
Early testing revealed technical hurdles that forced technology teams to extend evaluation timelines. An initial internal website chatbot retrieved outdated documentation buried deep within digital archives because search algorithms lacked human context. “That was, again, an example of we had to clean up that website in order to put a chatbot in front of it,” Hogan told National Mortgage News.
Hogan observed that traditional software operates as expected once it stops crashing, whereas generative models require extended behavioral refinement.
How business experts built custom agents
Pennymac shifted development responsibility from centralized technology departments to specialized business units. The lender assembled teams that included business people taken out of their day jobs to help automate tasks they know well. Hogan indicated that training mortgage professionals to build prompts proves more efficient than teaching software developers industry compliance. “You need someone who can understand the subtleties of the response,” Hogan told National Mortgage News.
Hogan indicated that for a given agent, such as one for closing disclosure review, it might take several weeks of testing. “But yeah, that time is, I think in the beginning they though, oh, we’re done.”
Where future skill sets are heading
Hogan advised technologists to master underlying business operations rather than relying entirely on programming proficiencies. Intellectual curiosity is now a primary asset for employees working in automated environments.
“Your coding knowledge is just not as special anymore, your business knowledge is,” Hogan told National Mortgage News.
Operational Takeaways for Mortgage Lenders
- Pennymac established isolated enterprise data spaces to eliminate third-party security exposure.
- Internal testing phases require two to four weeks of edge-case evaluation by domain experts.
Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.
