Prasad Kodibagkar, Chief Technology Officer at Polunsky Beitel Green, authored an article in National Mortgage Professional on where AI can genuinely help in the mortgage closing process and where it needs guardrails.
Kodibagkar argued that the closing stage leaves little room for error, since most of the loan process is already behind the borrower by that point. He pointed to document classification as one of the most practical near-term uses, helping reviewers reach the relevant sections of a large file faster without changing who is accountable for the review.
He also cautioned that any automation must be measured against what a trained person would produce. Because large language models generate probabilistic rather than fixed outputs, Kodibagkar said he looks for tools that constrain results to a defined structure, ground answers in source documents, and route uncertain cases to a human reviewer rather than a guess.
On agentic AI, Kodibagkar described it in operational terms, breaking a process into discrete steps handled and checked by separate agents, comparing it to expense report processing where one agent extracts data, another verifies it, and a third enters the result.
He closed on data governance, noting that any AI tool touching a loan file needs zero data retention, contractual commitments against training on borrower information, and a controlled environment rather than a consumer-grade tool.
Read the full article in National Mortgage Professional: What Mortgage Professionals Should Know About Practical AI in the Closing Process
