There are two quite different things sold under the name "invoice capture", and the pricing pages don't distinguish between them.
The first reads the top of the document: supplier, invoice number, date, total, tax. Four or five values. This is header-level capture, and it's what most built-in tools do — including the ones bundled free with accounting software.
The second reads the body: every line, with its description, quantity, unit price, line total and tax treatment. A twenty-line invoice produces a hundred values instead of five.
The gap between them is where the remaining work lives.
When the header is enough
Genuinely, sometimes it is.
If a supplier invoice always codes to one account — your accountant's monthly fee, a software subscription, the window cleaner — then the header gives you everything. Supplier, amount, date, done. Line detail would be noise.
For a business whose purchasing is mostly like this, header capture solves the problem and anything more is over-engineering.
When it isn't
It stops being enough the moment one invoice needs splitting across more than one account.
A builders' merchant invoice for £1,248 might be timber, board, fixings, insulation and delivery. Those don't all belong in the same nominal code. Materials for one job, consumables for another, delivery somewhere else again.
Header capture gives you one figure — £1,248 — and you open the PDF and type the breakdown in by hand. The tool has saved you the supplier name and the date. You're still doing the part that takes the time.
This is also true for:
- Job or project costing. You need line detail to attribute costs to the right job.
- Mixed VAT rates on one document. Standard-rated goods and zero-rated items on the same invoice can't be resolved from a header total.
- Anything you re-bill. If you pass costs to a client, you need the lines.
- Stock. Quantities live on the lines and nowhere else.
- Matching against a purchase order. A short delivery invoiced in full is only visible line by line.
Why line items are harder
Worth understanding, because it explains why the split exists.
Header fields sit in predictable places. The total is the largest currency figure, usually bottom right. The invoice number is near the top, labelled. There's a limited set of layouts and the values are individually findable.
Line items are a table, and tables are the hard problem in document extraction. The tool has to work out where the table starts and stops, which columns mean what, whether a description wraps onto a second line, whether an indented row is a sub-item or a new one, and what to do when the table breaks across a page — where the header row may or may not repeat.
Get the column mapping wrong by one and every quantity becomes a unit price — a failure a blended accuracy score will not show you.
That difficulty is the whole reason header capture is the default. It's the tractable part of the problem, and it is worth asking about directly before buying.
How to tell what you're being sold
The wording on marketing pages is slippery. "Extracts all your invoice data" can mean either. Some questions that get a straight answer:
Ask for a sample output on a multi-line invoice. Not a demo video — the actual extracted data from a document you send. It either has the lines or it doesn't.
Ask what happens when a table spans two pages. This is where line extraction most often falls over.
Ask whether line descriptions come through in full. Some tools truncate to the first few words, which is useless for coding.
Ask if lines can carry their own tax treatment. A single VAT rate applied to the whole document defeats the purpose.
The test that matters
Take your most awkward supplier invoice. Not a clean one — the one with twenty lines, mixed rates, and a description column that wraps.
Send it to whichever tool you're evaluating. Look at what comes back.
If you can post the result without opening the original PDF, the tool has done the job. If you still need the PDF next to you while you type, it's read the header and left you the rest.
Cribble reads to the line item — descriptions, quantities, unit prices and per-line tax — across seven document types. If you want to see it against your worst multi-line invoice, send it over.