Sometimes the OCR will group things together that are too long.For example, sometime the date will have extra numbers before or after. This is where a human eye can be better than a machine.
This can be due to bad printing, poor spacing, or machine error (the foil of machine learning).
The processing for Drag&Drop takes place in the "brain" of the DOKKA system, where the deep magic happens. New versions of the brain are put into production periodically, which improve this processing. It's not an exact science, but we are making it better with every version.
For now, unfortunately, either drag everything and delete the extra or type it by hand. For scans of printed receipts, keep trying the Drag&Drop, since it's possible that one will be bad and the rest perfect.