An AI review tool that extracts appraisal evidence and checks it against the source

Problem

From copying numbers to checking their source, the work was manual.

Construction appraisal means sorting hundreds or thousands of pages of receipts, payroll ledgers, attendance logs and bank statements by cost category, and checking that they agree.

Copying documents into tables took a long time, and verifying a number meant finding the source page again. Shift and surcharge calculations and comparing claims with actual payments were separate jobs.

What we did

We built the whole review flow, from extraction to cross-checks, into the system.

• Evidence extraction by cost category PDFs are split and analyzed, and AI extracts the items needed for 9 categories such as labor, electricity, utilities and travel.
• Each item linked to its source Every item carries a confidence score and page/position, so the review panel shows the exact PDF region.
• Review, totals and Excel in one place Extracted values can be edited on screen, with subtotals by appraisal period and per-category tables exported to Excel.
• Automatic shift and surcharge calculation Daily shifts per worker are read from payment statements, and night/holiday surcharges are calculated.
• Cross-checking attendance and payments Attendance is compared with claimed shifts, and transfers with net pay, flagging differences, unpaid items and transfers not on the statement.

Result

Less time organizing material, more focus on checking the evidence.

Uploading evidence produces per-category tables and cross-check results. Reviewers jump to each value's source, correct what's needed and export to Excel.

The system handles repetitive entry and calculation, so reviewers concentrate on flagged items and documents that need checking.