Take a photo of a receipt, or use my sample. Amazon Textract reads it and
sends back the vendor, the date, every line and the total, each with where it found it on
the page. The boxes you will see are drawn from that answer, on your own photo.
The scanner
One call to AnalyzeExpense per photo. Five reads per visitor a day.
Bed 01 · Textract
Place a receipt here. Flat, well lit, the whole slip in frame.
Nothing on the bed yet.
Your photo. It goes to Amazon Textract to be read and is not
stored anywhere. Before it leaves your browser it is shrunk to 1600 pixels on the long edge.
Crop out anything you would rather not send.
What comes back
Who sold it: name, address, phone, tax number
When: the receipt date
Money: subtotal, tax, total, amount paid
Each line: item, quantity, unit price, price
Everything else it could pair up, like pump or transaction number
Each field carries a confidence score and a box on the photo. Point
at a row to light up its box, or at a box to find its row.
Total—
Vendor—
Date—
Read in
—
Cost
—
Model
—
Fields
Field
Value
Confidence
Line items
Item
Qty
Each
Price
Confidence
Every field Textract returned (0)
Type
Label on the receipt
Value
Conf.
Where this fits
Paperwork in the field
This page is a working demo of Amazon Textract's expense analysis, AnalyzeExpense, at about a
cent a page. The same approach suits fuel receipts and delivery slips in a field app, so staff
don't retype them.
The confidence score is the part worth designing around. A field that comes back at 100 needs
little more than a glance; one that comes back lower is the one to put in front of a person. That keeps people checking numbers instead of keying them in.
If your team copies numbers off paper into a system, that is usually the cheapest hour in
the business to take back.