There isn't a method out there that doesn't have major flaws, ladies and gents. Thing is, many large, decentralized companies consider such surveys the easiest and most reliable way to measure customer satisfaction and outlet competence.
The fundamental balance is this: How does a company gather satisfaction data from hundreds of nodes (read: dealers) without making the collection so involved that nearly no consumers are willing to invest the time to fill the survey out? This is statistical survey/data analytics theory 101 kind of stuff.
Data is only as good as two major things:
1). Its volume
2). Its reflection of as large and even a cross section of the study group (customers) as possible
If you make the surveys too involved, you automatically skew the survey to the point of worthlessness because those with major complaints will be, by far, the largest portion of the subject field to fill them out. Yet if you make them too easy, you don't gather enough meaningful information to solve issues.
That's why many companies -- not just BMW -- actively promote the 'all 10s' request. It's a skew metric that basically equates to a data point that indicates satisfaction,
but not complete satisfaction -- and it's a far more valuable metric than surveys that don't ask deep enough questions or surveys that have an extremely low response rate. Algorithms exist that incorporate this skew effect into data analysis not as raw itemized multipoint data, but as its own data point.
Personally ... I don't fill them out. I'm too honest and critical for my own good.
