"Automate invoice data entry" sounds like one job. It is six, and only two of them are typing. Bookkeepers who automate the typing and leave the other four alone are often surprised that the month does not get noticeably shorter — because the typing was not where the hours were.

So this starts with the stages, in the order they actually happen, and which of them a machine can take.

The six stages of getting an invoice into the books

Collection. Finding the invoices. Three email inboxes, a supplier portal that emails a link rather than a document, the one supplier who still posts paper, and the director who forwards things a fortnight late.

Capture. Reading the supplier, reference, date, net, tax and total off the document. This is the stage everyone means by "data entry".

Coding. Deciding the nominal account, the cost centre, the project, the VAT treatment. This is judgement, and it is the stage that most often needs someone who knows the business.

Matching and approval. Checking the invoice against a purchase order or a delivery note, chasing whoever has to approve it, waiting.

Posting. Entering it into the accounting system, or importing it.

Filing. Storing the document somewhere it can be found again when an auditor or a supplier asks about it.

Before automating anything, time these separately for one afternoon's worth of invoices. Not estimated — timed. Almost everyone finds the distribution is not what they expected, and the stage that turns out to dominate is usually collection or chasing rather than capture. Automating the wrong stage is the most common way this project fails, and a single afternoon of measurement prevents it.

Capture is the stage a machine genuinely takes

Of the six, capture is the one that is pure transcription: the answer is printed on the document and the work is moving it into a grid. No judgement, no context, no decision.

It is also the stage that is most expensive to do by hand, and not because of the keystrokes. The cost is the concentration — the reason you cannot do it while being interrupted, and the reason it gets batched into an unpleasant block at month end rather than done as invoices arrive. Transcription errors are also the least detectable kind, because a mistyped total looks exactly like a typed one.

What capture produces is a sheet: one row per invoice, under headings you chose, with a column recording which file each row came from. Extracting invoice data to Excel goes through the field choices in detail, and invoice to Excel walks through the workflow end to end.

The columns to capture, and the two people leave out

Source file
Supplier
Invoice number
Invoice date
Due date
Currency
Net amount
Tax amount
Gross amount
Purchase order reference

Source file is the one to insist on. It is what turns the sheet from a list of numbers into something auditable: any row can be traced to the document behind it, by anyone, months later. Without it, a figure that looks wrong has to be found again from scratch.

Due date is the one people forget, and it is the one that makes the sheet useful for something other than posting. A captured due date gives you a payment run, an ageing, and an answer to "what is going out this week" that does not require reopening anything.

Two rules about the rest. Keep currency in its own column, never inside the amount — an invoice silently converted at an unrecorded rate cannot be reconciled against the payment that settles it. And leave a field empty when the document does not state it, rather than filling in something plausible; a blank is a fact about the invoice, and a filter on blanks is how you find the invoices that are missing a PO reference.

Code by rule, review the exceptions

Coding is judgement, so it cannot be automated wholesale — but most of it is not really judgement, it is repetition of a decision someone already made.

The practical approach is a supplier-to-nominal table, maintained deliberately. Most suppliers always code the same way. Apply that by lookup, and spend your attention on the rows the lookup cannot answer: a new supplier, a supplier who sells you two different kinds of thing, an invoice whose amount is well outside that supplier's usual range.

That last one is worth building in as a check rather than a habit. Sort by supplier, then by amount, and the unusual invoices are adjacent to their own normal ones. Most coding errors and most duplicate invoices are visible from that one sort.

Duplicates deserve their own check, because they are the error that costs actual money: supplier plus invoice number should be unique across the period. One conditional format finds every breach.

Match before you post, not after

If your invoices relate to purchase orders, capture the PO reference as a column and do the matching in the spreadsheet, before anything reaches the accounting system. Getting both sides into one sheet is the whole of the technique — extracting purchase orders to Excel is about exactly that.

Matching after posting means corrections, and corrections in an accounting system are more work and more visible than a column of mismatches in a sheet nobody else has seen yet.

Post by import, and reconcile the import

Most accounting systems accept a CSV or XLSX import of purchase invoices. If yours does, the captured sheet is most of the import file already: rename the columns to the system's template, and keep the original sheet alongside it.

Two specifics that save a bad afternoon. Reference numbers with leading zeros and dates in a non-US format are the two things CSV loses on import — PDF to CSV vs PDF to Excel explains why, and why XLSX is safer when the system accepts it. And reconcile after importing: the sum of the gross column in your sheet should equal the sum of what the system now holds for that batch. A control total on every batch is the cheapest audit habit there is.

What not to automate

Approval. A workflow tool can route and remind. It cannot decide, and an approval nobody actually looked at is worse than a slow one.

Coding you have not yet made a rule for. Automating an unexamined decision multiplies it.

A supplier sending one invoice a quarter. Automation has a setup cost and a maintenance cost. Four invoices a year does not repay either.

The first month. Run the automated capture and your existing process side by side for one period and compare the two sheets. It is the only way to find out what your documents actually do, as opposed to what you assume they do, and the disagreements are informative in both directions.

Start with collection, even though it is boring

If the timing exercise above says collection dominates — and it often does — the highest-value change has nothing to do with reading documents.

One destination for every invoice. A single mailbox or folder that all suppliers and all colleagues send to. Most of the chasing in accounts payable is chasing documents, not approvals.

Tell suppliers to send PDFs, not scans. A supplier emailing a scan of a printout of a PDF almost always has the PDF. One email to their accounts department removes an entire category of problem — see converting scanned PDFs to Excel for what scans cost you downstream.

A consistent filename convention. YYYY-MM-DD-supplier-reference.pdf sorts chronologically, searches usefully and makes the Source file column immediately meaningful.

Check what you have before you start. PDF Page Counter counts pages across a whole folder in the browser, which is a quick way to spot the statement that arrived instead of an invoice, and Extract Text from PDF tells you which documents have a text layer and which are scans — two different amounts of work that are better separated than mixed.

Common questions

Will this replace a bookkeeper? No. It removes the transcription and leaves the coding, the matching, the judgement and the conversations — which is the work that needed a bookkeeper in the first place.

Do I need to set up a template per supplier? Not with the approach described across this blog. You name the fields you want by meaning rather than mapping positions on a layout, so a supplier you have never dealt with needs no setup. Extracting specific fields from PDFs is the detail.

What about invoices with line-item detail? A different output shape — many rows per document rather than one, with the header fields repeated onto each row. Extracting invoice line items sets out the layouts and the reconciliation check.

How do I handle credit notes? In the same sheet, as negative values, with a column marking the document type. Keeping them separately means every total has to be computed twice, as consolidating invoices into Excel explains.

Should I automate everything at once? No. Capture first, because it is the stage with a clean boundary and a verifiable output. Then whichever stage your timing exercise says is next.

The order that works

Measure the stages. Automate capture. Put coding behind a supplier lookup and review the exceptions. Match in the sheet before posting. Reconcile every batch against a control total.

Then measure again, because after capture is gone the bottleneck has moved, and it is rarely where it was. How it works describes the capture step as this site implements it.