What the AI does
- 01
Document check
Reads every document as it arrives, pulls out the key details, checks them against what the deal needs and what the client declared, and lists what is missing. Each gap goes on one follow-up list.
- 02
Credit write-up
Drafts the credit assessment from the file: security, servicing, exit and risks.
- 03
Legal check
Reviews the security and the legal risk in the file.
- 04
Second opinion
Argues against the credit assessment, hunting for weak spots, before the Univo credit team reviews it.
What it produces
Together, these read every document on a file and draft an assessment: what the documents show, what is missing and what does not match the declaration. The credit assessment is Univo’s own credit work, and it starts by itself when a deal is submitted.
The AI also fills in the application for the broker to check, from the documents and a point-form summary. Anything it takes from a document is checked against the document itself, and nothing is saved until the broker applies it.
Where the documents are read
Credit assessment uses AI models that run on our own server. Application documents are not sent to an outside AI service to be assessed.
Who decides
A program compares each deal with lenders’ criteria. No program approves, declines or prices a deal.
After the second AI check, people on the Univo credit team review the assessment and decide. Approvals and credit decisions are staff-only actions, recorded against the person who made them.
The one cloud path
The in-portal question assistant for brokers and staff is the exception. It can send a summary of recent deals (borrower name and type, loan amount and status) and the typed question to Anthropic in the United States, and OpenAI checks its draft answers. Documents and their contents are never sent on this path, and nothing from the lender or legal portals is.