And I was, conversely, not talking about what actually happens by default, but rather what way you need to force yourself to think if you want to avoid bias when doing something important involving the corporate reputations of others—like, say, picking stocks.

If you have both information about a "brand", and information about the actual group or individuals underlying it, giving any weight to the brand will give a suboptimal result and let people exploit you. The reputational information has some value, but only if you have no "observational" information. Trying to use them both together is double-counting. Just use the information about the individuals and throw the brand information away.

In in other words, basically http://lesswrong.com/lw/lx/argument_screens_off_authority/, but with a more generalized halo effect in place of "authority."