The last resort is measuring a print to scale dimensions from. I can sometimes import a PDF, if supplied, into CAD software and 'trace' over it to get pretty close. Usually I'll have enough dimensioned data to align and scale the print to meet the lines I've drawn from dimensioned data. That will get the undimensioned features on the imported PDF (sometimes a scanned file) located NEAR-TO-TRUE size, location, and orientation.
Knowing that typically people will not go out of their way to place things in oddball locations, I might round things to the nearest 1/32" or center things that look like they should be centered, etc. So long as it suits the assumed/known function of the part at hand. Many times assumptions can safely be made if you know what the thing IS.
When you cannot come to an agreement with the customer, there's been two successful approaches IME. One is, time allowing, sending them a sample piece as a 'first article' to approve along with a drawing we create, and all subsequent parts will be made to that 'shop drawing' and supplied sample part. Another is supplying our 'best effort' on the minimum # of parts the customer needs to 'get by' for now, and then making any revisions to our 'shop drawing' based upon feedback from our initial parts. This is based on the assumption that the customer will be paying for all of those 'best effort' tryout pieces, or understanding that it may come with some modifications as necessary in-use.
A lot is based on the relationship and communication with the guy holding the check book.