Solusion to discover/inference the usage/meanings of tables in unkown database?

This is a usual situation I meet recently that customers gave me a database with many tables they don't quite understand too, then ask me to make a model predict the future revenue, classify which user may be valuable or something else.

To be honest, extracting useful data from an unknown database made me exhausted.

For example, I need to figure out

1. which table is the user table, product table, or transaction table ...
2. which column can use to join(there may not have the foreign key)
3. what relationship may exist between two tables

I have thought user_id may have features:

1. is unique in user relative table, but can be duplicated in transactions
2. length may lower than a value
3. have one to many relationship between user table and transaction table.

But this prior information is poor, lack of sample is another problem.

I searched for competitions and paper but nothing was found.