AI Smart Matching

 🚧 Early Access / Beta: AI Smart Matching is currently available to selected customers only and carries a Beta label as we continuously refine its capabilities. Contact your FileInvite account manager to inquire about access. 

How AI Smart Matching Works

When a document is uploaded to the Client Portal (or when a lender pulls files in on behalf of a borrower to reduce workflow toil):

1. Content Analysis: FileInvite’s AI looks inside the document's content to analyze the text, layout, and keywords (such as identifying fields that signify a passport or a bank statement). Note: The AI excludes the file name entirely from its matching prompt, as file names are often unreliable markers.
2. Contextual Matching: The AI evaluates the Request Name and Request Description to understand what document you are seeking. It then matches the document content to the most appropriate request block.
3. Contact Attribution: On Invites featuring multiple contacts, the AI securely scans for identifying information (such as the borrower's name) to automatically attribute the file to the correct person.
4. Automated Routing: The file is immediately routed into the corresponding request row, flagged and ready for lender review.

No initial action is required from the borrower or your team for matching to occur.

Supported File Types

The AI engine supports a broad range of standard file families commonly used in lending workflows, including:

• Documents: PDF, CSV
• Images: JPEG, PNG (including clear screenshots of official documents, like a smartphone screenshot of a utility bill)

Note: Non-standard or unusual file families (such as raw HTML files) are not supported.

Reviewing Matched Documents & Handling Exceptions

• Instant Visibility: Matched documents appear in the correct request row, attributed to the correct contact, as soon as they are successfully processed.
• Standard Approval Workflow: Review, approve, or reject matched documents the same way you would any other file directly from the Invite view.
• Manual Re-assignment: If a document is routed to the wrong request row, you can easily shift it to the correct request manually.
• Handling Unmatched Files (Exceptions): Documents that cannot be matched with sufficient confidence, or files that don't match any active requests on the Invite, are flagged as unmatched in the side drawer interface.
  • Lenders have the flexibility to manually assign these unmatched files to a request.
  • Any extraneous or unassigned files left over at the end of the session are automatically deleted to keep your workspace secure and clean.

🔒 Privacy, Security, and Data Use

We adhere to a strict "No Phone Home" policy regarding data security and artificial intelligence:

• No Model Training: FileInvite never uses your document content, borrower PII (Personally Identifiable Information), or private text to train or fine-tune public AI models.
• Secure Infrastructure: All document analysis takes place entirely within a highly secure AWS Bedrock environment. Customer data remains isolated within this walled infrastructure and is never transmitted out to public or foundational AI model providers (like OpenAI or Anthropic).
• Data Isolation: Borrowers' data remains entirely private and is used solely for the isolated transaction of classifying and routing files within your active Invite.
• Product Analytics: To continuously monitor system performance and control costs, FileInvite only tracks anonymous structural metadata (e.g., whether a match succeeded or failed, token consumption volumes, and broad document classifications like "bank statement vs. personal ID").

Important Technical Notes

• Activation: AI Smart Matching must be enabled at the account level by your FileInvite account manager. Once flipped on, the button interface in your sent invites will transition to read Smart Matching. It will automatically apply to all new Invites.
• No Retroactive Matching: Documents uploaded before AI Smart Matching was enabled on your account will not be retroactively analyzed or sorted.
• No Structural Validation: Smart Matching is built purely to organize and sort files to eliminate manual document classification. It does not validate or audit the legitimacy of the file's data (e.g., verifying if a bank statement is authentic or altered).
• Evolving Architecture: To preserve flexibility, optimize processing speed, and control token costs, FileInvite utilizes various evolving models within AWS Bedrock (such as Anthropic Claude Sonnet). Because models evolve rapidly to improve accuracy, we do not commit the system to a single fixed provider or a static minimum accuracy rate.

💡 Best Practices:

• Write specific, unambiguous request names — these are the primary signal the AI uses to match incoming documents
• Review matched documents promptly and reject any that have been incorrectly routed to keep Invite status accurate
• If you notice consistent mismatches on a specific request type, contact FileInvite support