Where the Time Goes in Accounts Payable
An AP clerk's day looks much the same at most companies. Open the supplier's statement as a PDF or scan, find the matching purchase order in the ERP, check item, quantity, and unit price against the goods receipt line by line, then key in the entry. The same information gets read by eye at least three times and typed by hand once more.
At 3,000 documents a month and an average of six minutes each, that adds up to 300 hours a month, roughly two full-time people doing nothing but matching and data entry. The bigger problem is how errors spread:
From Reading Documents to Extracting Fields
Every supplier formats its statements differently. Plain OCR, which only reads characters, isn't enough. You need field extraction that can tell whether a number is a unit price or a line amount.
Comparing Extraction Approaches
Don't Run Recognition on Documents That Are Already Data
In Korea, e-tax invoices can be received as structured data through the National Tax Service system or an e-invoicing service provider. Converting them back into images and running recognition on them only lowers accuracy. Bring in data-native documents through integration, and save recognition for documents that have no data source, such as paper or PDF supplier statements.
Three-Way Match: Rules for Reconciling PO, Receipt, and Invoice
A three-way match reconciles the purchase order (what you agreed to buy, at what price), the goods receipt (what actually arrived), and the invoice (what you're being billed).
Tolerances and Partial Deliveries
Classifying Exceptions and Routing Them
You can only send an exception to the right person once you know why it failed:
Send People Only What Needs a Human
The goal isn't to auto-approve everything. It's to auto-approve what you can trust and send only the uncertain cases to people.
The review screen should show the original image and the extracted values side by side, and clicking a value should highlight where it came from on the page. If only low-confidence fields are color-flagged, reviewers don't have to re-read the whole document. Each correction they make becomes training data for the next extraction.
Metrics That Show Whether It's Working
The biggest driver of automation rates is often supplier master data quality, not model performance. If one supplier is registered two or three times under slightly different names, or there's no mapping between supplier item names and your own item codes, documents that were extracted correctly still fail at the matching step. Automation projects often have to start with a master data cleanup.
How POLYGLOTSOFT Builds AP Automation
POLYGLOTSOFT builds AP automation without replacing your existing ERP. We add a layer in front of it for document intake, field extraction, three-way matching, and review. You can phase it in: e-tax invoice integration, supplier statement extraction, supplier and item master cleanup, and finally automatic posting of approved entries to the ERP. If invoice processing is tying up your team, let's start by mapping how your AP workflow runs today.