How to

Scans and phone photos work too

A scanned PDF is a picture of a page, and so is a phone photo: there's no text to copy out. Mammoth reads the image, so you upload them like any other file and check the rows before saving.

53 seconds, real app, made-up invoice and receipt. Follow along with the same two: Scanned invoice (PDF, no text layer) and Receipt photo (JPG). Download the video (MP4)

The steps

1
Upload the scan
In Data, choose New dataset and upload the scanned PDF, then choose Extract Data. You can't select any text in this PDF: it's an image of the page.
2
Read the table
Choose Only tables, then Extract. Mammoth finds the invoice lines: item, quantity, unit price and amount, 5 rows. Compare them with the scan on the left, then choose Looks right and Create dataset.
3
Upload a photo
Phone photos go in the same way: JPG and PNG files upload like any other. This receipt was photographed at an angle and has no table lines.
4
Say what you want from it
With no table to find, choose Custom and describe the rows: one row per item, with shop, date, receipt number, item and amount.
5
Check the rows
4 rows, one per item, with the shop, date and receipt number on each. Check the amounts against the photo before you save.

For the best read

Mammoth reads what it can see, so a clearer image gives a better result.

Photograph the page flat, in good light, with all four edges in the frame
Scan at a readable resolution; a blurry or low-resolution scan reads poorly
For receipts and forms, use Custom and list the fields you want

Whatever you upload, compare a total or a few values with the page before you save.

If it doesn’t look right

A value was misread: Check the image is sharp where that value is. Rescan or retake the photo if it isn't, and upload it again.
The header details are missing: Only tables keeps the table rows. To get the invoice number or date on each row too, choose Custom and list them with the line items.
It takes a while: Scanned documents take longest, multi-page ones especially. If it runs long, reopen the file in a few minutes to see the result.