Skip to content
All the answers
Answers

How do you automate invoice data entry?

You photograph or forward the document, a model reads the fields off it, and the numbers get checked against something other than the model's own confidence. The reading is the easy half. What makes a pile of documents safe to file is being told which ones came out wrong, and most tools are quiet about exactly that.

Last checked 14 August 2026. Prices on this page are ours and current.

The five steps, and where each one breaks

First the image is prepared. Resized, straightened, and measured for focus. This matters more than it sounds: a photograph too soft to read is the single commonest reason a number comes out wrong, and it is detectable before anything else happens.

Then the page is read. A model returns the supplier, the invoice number, the date, the totals and the line items. It also returns how sure it was of the characters, which is not the same as whether the numbers make sense.

Then the numbers are checked with arithmetic rather than opinion. Does the subtotal plus the tax equal the printed total? If it does not, something is wrong on the page or in the reading, and either way a person should look.

Then the date is resolved, which is where quiet errors live. 04/08/2026 is the fourth of August or the eighth of April depending on who wrote it, and nothing on most invoices settles it. A tool that silently picks one has filed a guess.

Then it goes somewhere: a spreadsheet, your accounting software, a searchable record you can come back to in six months.

Accuracy is the wrong question

Every vendor will tell you their accuracy is high, and on clean documents everybody's is. The number that matters is different: of the documents it got wrong, how many did it tell you about?

A tool that reads 98 out of 100 correctly and says nothing about the other two is worse than one that reads 95 and hands you the five. In the first case you have two wrong numbers in your books and no idea which. In the second you have five minutes of work and clean books.

So when you evaluate anything, hand it your worst documents, not your best. The blurred phone photo, the crumpled thermal receipt, the one where the supplier's own arithmetic is off. What you are testing is not whether it reads them. It is whether it admits it could not.

What it costs to run

Reading a page with a current model costs cents, not dollars. We charge $10 per 100 documents for the reading and the spreadsheet, and $20 per 100 if you also want every document kept and searchable afterwards. No monthly fee and no minimum.

That pricing works because the model call genuinely is cheap. If somebody quotes you a large monthly platform fee for the same job, ask what the fee is for. Sometimes the answer is good, such as an integration that saves a person a day a week. Sometimes it is margin.

Where it stops being about invoices

The reading is worth twenty seconds a document. The part a business actually feels comes later, when every document has been read and the supplier, the date and the total are fields rather than pixels.

At that point a question like everything between 30 and 500 dollars in July is a filter instead of an afternoon in a filing cabinet. A folder of photographs cannot answer it. That is the second price on our list rather than a separate product, because it is simply what you have left once the reading has happened.

Do not take our word for it.

Drop a folder in. It reads them in your browser and tells you which ones it would not trust.

Read your own invoices
Also asked

Does it work on photographs, or do I need scans?
Photographs are fine and are what most people actually have. Focus is the thing that matters, not the device: a steady phone photo reads better than a rushed scan.
What about handwriting?
Printed figures read reliably. Handwriting varies enormously and is the case where you should test your own documents rather than trust anyone's demo, including ours.
Can it read PDFs with several invoices in one file?
Yes, though a page per document is cleaner. Tell us what your files actually look like, because that shapes the sensible answer more than the software does.
Where does the data end up?
A spreadsheet by default, since everyone can open one. We also write files that accounting software imports directly, and we can keep a searchable record of every document read.

Tell us what you are trying to work out and we will answer it straight, including when the answer is that you do not need us.

Or email hello@ahlelak.com