What we do
Tools
Industries
Department
How we workLet's talk →
Home/Invoice Processing Automation
Invoice processing

Automated Invoice Processing: Where Invoice OCR Stops and a Person Takes Over

Capture tools read the tidy PDF from a vendor who hasn't changed template in years. The scanned fax, the forty-line distributor invoice and the vendor who redesigns every quarter are what still land on someone's desk. This page is about that intake step: reading, checking and matching. For approvals and payment, see our accounts payable automation page.

Where a month of invoices actually goes

Most teams with a capture tool still touch a big share of their invoices. The aim is to shrink the bottom line to the ones that deserve a person.

Example month, 400 vendor invoices
Invoices arriving by email, vendor portal and post 400
Clean PDFs, read and matched to a PO, posted untouched −240
Duplicates caught before they post −10
Still reaching a person (scans, long line items, missing PO numbers, prices off the PO) 150

Illustrative numbers. Your own split depends on how many vendors you have and how they send invoices, which is the first thing we look at.

What capture does with each kind of invoice

What arrives What capture tools handle now Where it still needs work
A clean PDF from a regular vendor Header fields and totals, read reliably Line items. Many setups only capture the total, so coding by line still happens by hand.
A scan, fax or phone photo of a paper invoice Partly. Confidence drops with skew, stamps and handwriting Low-confidence fields posted as if they were right. They need flagging, not a guess.
A long distributor invoice, dozens of lines over several pages Varies a lot between tools Lines split across page breaks, and subtotals read as extra lines.
A spreadsheet, CSV or statement instead of an invoice Usually not at all These need their own import, mapped to your vendor and account codes.
An invoice in the email body, or behind a download link Often missed, because there is no attachment to read Fetching the document from the link, or reading the email text itself, before capture runs.
One vendor billing several of your entities or locations Vendor found, entity often wrong Ship-to and bill-to details deciding which set of books the invoice posts to.

Read correctly, then stuck at the PO match

Capture is only half of intake. These are the matching problems that hold up invoices the OCR got right.

No PO number on the invoice, or the wrong one

How you'll know: Invoices sit unmatched until someone searches open POs by vendor and amount.

Fix: Look up open POs for that vendor and suggest the likely match, so a person confirms instead of searching.

Prices a few cents off the PO

How you'll know: Rounding and small price changes send every second invoice to approval.

Fix: Agree a tolerance per vendor or per amount. Differences inside it clear, and anything outside it goes to the buyer with both prices side by side.

Partly received orders

How you'll know: The vendor bills the full PO while the warehouse has only logged part of it.

Fix: Match against what was actually received and hold the rest until the receipt arrives.

Freight, fuel and surcharge lines that were never on the PO

How you'll know: A perfect match fails on one extra line.

Fix: Code those lines by rule to the right accounts, within a limit, rather than failing the whole invoice.

Cases on the PO, eaches on the invoice

How you'll know: Quantities look wildly wrong although the order was fine.

Fix: Store the unit conversion for each item so the match compares like with like.

What n-frames builds into your invoice intake

01

Capture tuned to the vendors you really have

Extraction set up against the formats your vendors send, rather than the tidy sample invoice from a sales demo.

02

Exceptions flagged instead of miscoded

An invoice the capture tool can't read with confidence goes to a person with the doubtful fields highlighted, never posted with a guessed figure.

03

Coding that follows your chart of accounts

Lines coded to the right GL account and department from vendor and item history, so nobody overrides the same default every week.

04

Duplicate checks beyond the exact match

Each invoice compared with what's already been processed, including near-duplicates: same vendor and amount under a different invoice number.

05

Straight-through posting for the clean ones

An invoice that matches its PO and passes every check posts to your accounting system with nobody touching it, so review time goes where it's needed.

Faster reading isn't the win. The win is a queue that only holds invoices with a real question attached, while the rest post correctly without anyone opening them.

What this looks like in practice

Field Sales Organization

Expensify for reps who travel constantly.

The problem Finance reads every expense report to check it against the policy and the trip budget, a job that grows with every new hire.

Reports are checked against policy and budget before they reach an approver. Finance reviews the flagged ones and the rest move on.
Let's talk

A pile of invoices your capture tool gave up on?

n-frames builds the intake around the tool you have: capture tuned to your vendors, PO matching with sensible tolerances, and a review queue that only holds the invoices a person really needs to see. A sample of your awkward invoices is where every one of these projects begins.

Shrink the review queue →

Questions people ask

How is this different from accounts payable automation?

AP automation covers the whole cycle from invoice to payment: capture, approval, matching and posting. This page is about the first part, getting accurate data out of whatever invoice turns up and matching it to the PO, which is usually where AP automation breaks first. The full cycle is on our accounts payable automation page.

What if our vendors all send different formats?

That's normal. We start from a sample of what actually comes in, PDFs, scans, spreadsheets and email-body invoices, and sort it into what can post on its own and what should still go to a person.

Does this replace our existing capture tool?

Usually not. The tool you have is often fine on the clean invoices. What's missing tends to be exception handling, matching and the link to your accounting system, so we build around it where that makes sense.

How do you handle a tool misreading an invoice?

Every extracted field carries a confidence score. Below the threshold, the invoice goes to a person with those fields marked, instead of posting a wrong number. Getting that cut-off right per vendor is most of the engineering.

Is the quote for invoice processing automation really free, with no obligation?

Yes. n-frames designs the intake against a sample of your awkward invoices first and prices from what that sample shows. You owe nothing for it.

Let's talk

Tell us the one thing your team does manually that eats up time. We read every message and reply within a day.