The useful data is usually already in the company: the CRM, the inbox, the spreadsheet someone updates by hand.
A new lake does not fix that. Neither does a model that has never seen how the work actually moves.
We start with the records people already trust, even if they are ugly. Then we decide what is worth connecting.
Clean data is a result of use, not a prerequisite you wait for.