Accounts payable · Production

UiPath Document Understanding automates invoice processing for major U.S. retailer, cutting per-invoice handling from 3-5 minutes to 30 seconds

The problem

A large U.S. wholesale retailer's AP team received 200–500 invoices per day from gasoline and freight vendors, peaking at 700, and each invoice required 3–5 minutes of manual work: opening emails, looking up supplier IDs, extracting data, and keying into the accounting system.

Workflow diagram · grounded in source
1
Bot reads invoice emails
trigger
“A bot reads emails containing invoices. If the email contains one or more attached files that are not invoices, the bot flags the email, which is then forwarded to an AP department staff member for manual evaluation. If the email only co…”
2
OCR digitization and page split
ai_action
“The bot digitizes the invoice using OCR to find and determine the locations of the header and footer for each page of the invoice. The bot then splits the invoice into its individual pages, with each page displayed in correct order.”
3
ML data extraction
ai_action
“The bot uses OCR and machine learning to extract information such as invoice date, number, amount, and due date. Accelirate used machine learning and AI Center to train the out-of-the-box model for invoices on the numerous formats of inv…”
4
Confidence-threshold routing
routing
“If the confidence level for the extracted data is above the 95% threshold, the invoice is sent automatically to a reconciliation queue for payment processing. If the confidence score on the information is below the threshold, it is sent …”
5
Human review at Validation Station
human_review
“an AP team member uses the UiPath Validation Station to inspect the invoice”
6
Continuous model retraining
feedback_loop
“Confirmed information from Validation Station is uploaded to AI Center to help continuously retrain models, making them more efficient as more data is fed to it. the machine learning done with UiPath Document Understanding includes a fee…”
7
Bot generates processing report
output
“the bot generates a report detailing count of items in each queue, how long the bot ran, and how many items were processed. All reports are archived for record keeping.”
Reported outcome

After deploying UiPath Document Understanding with OCR and ML, 93% of invoices now process straight through to the reconciliation queue at a 95% confidence score, and per-invoice processing time dropped from up to five minutes to about 30 seconds.
Accelirate estimates at least 20% of AP department staff can be redirected to higher-value work.

Reported metrics
Invoice processing time (old)three to five minutes per invoice
Invoice processing time (new)about 30 seconds
Straight-through processing rate93%
ML model confidence score95%
Show all 7 reported metrics
Invoice processing time (old)three to five minutes per invoice
Invoice processing time (new)about 30 seconds
Straight-through processing rate93%
ML model confidence score95%
AP department staff redirectable to higher-value workat least 20%
Daily invoice volume (average)200 to 500 per day
Daily invoice volume (peak)700
Reported stack
UiPath Document UnderstandingUiPath AI CenterUiPath Validation StationOCRRPA
Source
https://www.uipath.com/resources/automation-case-studies/major-retailer-rpa
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After deploying UiPath Document Understanding with OCR and ML, 93% of invoices now process straight through to the reconciliation queue at a 95% confidence score, and per-invoice processing time dropped from up to fiv…

What tools did this team use?

UiPath Document Understanding, UiPath AI Center, UiPath Validation Station, OCR, RPA.

What results were reported?

Invoice processing time (old): three to five minutes per invoice; Invoice processing time (new): about 30 seconds; Straight-through processing rate: 93%; ML model confidence score: 95% (source-reported, not independently verified).

How is this accounts payable AI workflow structured?

Bot reads invoice emails → OCR digitization and page split → ML data extraction → Confidence-threshold routing → Human review at Validation Station → Continuous model retraining → Bot generates processing report.

Related accounts payable cases