Modernizing San Francisco’s Invoice Review Process

The DxHub team created an intelligent reconciliation tool to assist the city’s current review process. Staff upload all documentation—invoices, payroll records, receipts, bank statements—in a single batch via drag-and-drop. The system automatically categorizes each document by budget line using filename conventions, with AI stepping in when files are ambiguous. 

The tool parses line items from CSV invoice files and matches them to uploaded PDFs using a dual-verification approach: amounts must match and at least one keyword must align. This prevents false matches that occurred in early testing with common dollar figures. When reviewers click on a sub-item, the system highlights the corresponding text directly on the relevant PDF pages. 

A Boolean search function allows users to query uploaded PDF documents using OR/AND logic with specific amounts and keywords. Reviewers mark each match as either “Invoice” or “Proof of Payment,” and the system generates a report showing which line items have been matched with corresponding backup documents. 

When automated matching falls short, reviewers can manually highlight text or link PDF pages to specific line items—preserving human oversight of the reconciliation process. As one client remarked after a recent demonstration, the tool was “a whole leap beyond what you showed us last time.”