Records that survive contact with reality.
The first deliverable is a count, not a file. Where a rule rejects a row, the reason travels with the rejection. Anything that could not be filled is stated plainly rather than padded.
Phone formatting
Rendering numbers in a consistent international form, with the original retained.
Franchise handling
Separating independently owned sites from company-owned ones where the evidence allows.
Syntax screening
Removing malformed and obviously generated addresses before anything is paid to check them.
Name the one field that matters most and we will start there.