DL and Bank Statements

I wanted to see what everyone’s thoughts were on using CNNs to parse bank statements.

The input: arbitrary number of pages (PDFs most likely) from any financial institution (from large banks to the smallest credit unions)
The output: A CSV which pulls out each transaction date, description, and amount.

Is this a tractable problem? Are there any similar uses for CNNs which I can use as a reference in developing this?

I reckon it’s the same problem like MNIST. Instead of 10 classes we would have more - alphabets, numbers, special characters etc.

And also there are many OCRs available, you can get pretty good results using OpenCV.