This morning a client moved their invoice OCR to Amazon Textract. I spent an extra 40 minutes fixing column shifts in the CSV export instead of keying — feels like more QA than entry, anyone else?