Invoice data is discussed as if it were one thing and behaves as three. Header fields come off the document reliably. Line items come off unevenly and are what the match needs. And the coding is not on the document at all, because it is a fact about your organisation rather than about the supply. Treating all three the same is where capture projects go wrong.
Reliable: the header
Supplier, invoice number, dates, currency, tax and totals. These appear in predictable places in predictable formats and modern extraction handles them well enough that checking each one individually is usually wasted effort. Spot-check them, and spend the saved attention on the fields that are actually uncertain.
Uneven: the lines
Descriptions vary between what you ordered and what the supplier calls it. Units of measure differ. One ordered line often arrives as several invoiced lines. Line extraction is what separates products, and it is what the three-way match depends on, so it is the thing to test on a genuine multi-line invoice from your own pile.
Absent: the coding
The general ledger account and the cost centre are judgements about your organisation, not facts printed on the invoice. Software can suggest from the supplier and from history, and a suggestion accepted without thought propagates last year's mistake. The reliable fix is capturing the coding at the order, where somebody actually knows.
Questions people ask about invoice data
What should we validate?
Supplier exists, invoice number is new for them, lines sum to net, net plus tax equals total, date is plausible, bank details match what you hold.
Should we store the original document?
Always, linked to the record. Extraction is an interpretation; the original is the evidence, and retention attaches to it.
Does structured invoicing avoid all this?
For suppliers who can send it, yes: the data arrives as fields with nothing to read and nothing to misread.
How much of this should we store?
All of it, alongside the original document. Storage is cheap and the fields you did not keep are the ones somebody asks about later, usually during a dispute or an audit when reconstructing them is expensive.
What is the single most valuable field?
The purchase order reference, because it is what connects the invoice to the order and the receipt. Without it every other field is accurate information about a claim nobody can check automatically, and the invoice takes the expensive manual route.