Your project is one type of what are called "establishment surveys". The basic sampling design is a cluster sample: you randomly select subsidiaries; then get enough information from each selected subsidiary to sample within it.
Although could take a simple random sample of subsidiaries, that could lead to sub-optimal precision if the number of line managers varies greatly among them. It pays to get some quantitative "size measure" that might correlate with the number of managers. Then you can sample with probability proportional to that size (PPS); or stratify by size and do simple random sampling within strata.
If there is no information, you could do a two phase sample: select a larger number of subsidiaries then you need; contact just them to get information on the number of managers; stratify the initial sample based on this information; and take a smaller sample of subsidiaries for interviews. Note that this design requires special adjustment to the standard errors. I believe that Tom Lumley's Survey package in R will do the correct analysis. If you attain good cooperation with a single subsidiary--necessary for pretests and pilot tests--you might find that approximate information for all subsidiaries is available within the company.
There are excellent case studies of sampling without initial lists in WE Deming, Sample Design in Business Research, Wiley, 1960.
It is difficult to write good questionnaires for these kinds of surveys and especially difficult to get cooperation and high response rates. For suggestions on many topics, see a recent handbook on business surveys: http://www.amazon.com/Designing-Conducting-Business-Surveys-Snijkers/dp/047090304X.