01 · The challenge
Credits were hiding in statements nobody had time to read
The company handles thousands of vendor statements every month. An earlier automation, built by another provider, could read only simple text-based PDFs, so most statements still needed manual review. Scanned and non-native PDFs failed OCR and credits were missed. Two to three people sifted partial OCR output and emailed vendors one at a time, and many statements were never reviewed at all, delaying credit recovery.
02 · The approach
A robot that reads every statement and drafts the follow-up
TSP built an AP credit-recovery automation that works in a loop with the AP analysts.
How it works
Automated collection
The robot downloads statements from Basware every day.
Intelligent OCR
UiPath Document Understanding reads PDFs of 1 to 100 pages, including scans, and Excel files.
Credit logic
Recognizes leading and trailing negatives, parentheses, dedicated credit columns and other vendor-specific patterns.
Data layer
Results are written to a SQL repository, with a refreshed master sheet after every run.
The follow-up
Analysts mark the rows that need outreach, and the robot generates tailored emails to vendors and restarts the cycle.
03 · The results
More than $40 million back in nine months
- More than $40 million recovered in the first nine months
- Statements processed up from 300 to 3,000 a month, a 900% increase
- Manual hours down from 40 to 60 a week to under 5, a 90% reduction
- Analysts now focus on complex discrepancies
- Faster vendor communication has improved relationships and shortened the time to receive credits
- New statement formats handled without code changes
