Solutions · Finance
Invoice Automation
Invoices read field by field, validated and matched before anyone touches them.
What comes off the page
Fields, Not Text
OCR gives you characters. What a finance system needs is a vendor, a date, forty priced lines and a tax breakdown that adds up.
Header
Invoice identity
Number, dates, currency, PO reference and payment terms, read wherever the supplier chose to put them on the page.
Header
Vendor identity
Name, address, tax registration and bank details, resolved against your vendor master rather than taken on trust.
Lines
Line items
Description, quantity, unit price, discount and line total, read as a table rather than flattened into a wall of text.
Lines
Tax detail
Rate, taxable value and tax amount per line, including invoices that carry more than one rate.
Checks
Arithmetic
Line totals, subtotal, tax and grand total are recomputed and compared with what the document claims.
Checks
Confidence
Every field carries a score, so you decide what posts automatically and what a person looks at.
Confidence
A Score on Every Field
One headline accuracy figure tells you nothing actionable. A score per field tells you exactly which values to trust and which to look at.
One invoice, as the agent returns it · Illustrative example
Thirteen fields cleared the threshold and posted. One handwritten amendment did not, so that row alone went to review.
Table parsing
Rows read as rows, so a 40-line invoice stays 40 lines.
Threshold routing
You set the score below which a field goes to review.
Format drift
A supplier redesigning their invoice is flagged, not silently misread.
Inputs
However It Arrives
Suppliers send what suits them. The agent reads all of it.
Native or scanned
Scan
Flatbed or MFP output
Photo
Phone camera, on an angle
Email body
Text invoices with no attachment
XML / EDI
Structured supplier feeds
Custom
Your own schema and rules
FAQ
Questions Teams Ask
How accurate is the extraction?
Field-level accuracy typically sits between 95 and 98 percent on standard formats once the agent is tuned to your documents. The number that matters more is the confidence score on each field, because that is what decides whether a value posts automatically or waits for a person.
What happens when a supplier redesigns their invoice?
Nothing breaks. The agent reads document structure rather than fixed coordinates, so a moved total or a new layout is still found. Where a change genuinely reduces confidence, those fields drop into review and you are told which supplier changed.
Can it read line items that run across several pages?
Yes. Tables are followed across page breaks, continuation rows are joined to the right line, and the totals are checked against the full set rather than the last page only.
Is there an API?
Yes. Submit a document, poll for the result or receive it by webhook, and get structured JSON with a confidence score per field. The same extraction runs whether a document arrives through the API, the portal or a watched mailbox.
Keep reading
Related Capabilities
AP Automation
What happens to the data once it is off the page: match, approve, post.
Read more →GST & Compliance
Tax read per line, checked against the code, reconciled before filing.
Read more →DocPro
The document engine behind this, and every other document type it reads.
Read more →FinSuite
The full finance workflow this capability feeds.
Read more →DocPro Invoice Agent
The agent that reads each invoice field by field, with a confidence score on every value.
Read more →Next step
Send Us Your Worst Invoices
The creased scans and the forty-line ones. Those are the useful test, not the clean samples.