• The technique only gets more powerful the more cousins you know of who have tested. On the side of my family I have researched the most extensively (3000+ known cousins out to about the fifth-cousin level) there are at least sixty people who have tested on AncestryDNA. When you have a color-coded group that large, the best new leads become really obvious - if a new, unknown match only matches 3 cousins out of the 60, that's not necessarily very promising. But if they match 25 out of those 60, while not having a similarly strong pattern of matches to other sides of my family? Then we're in business.  
• For reasons I don't fully understand, on AncestryDNA the Ashkenazi endogamy seems to play much less of a role in clouding the matches of people who are only part-Jewish. I am 100% Jewish, and have a handful of matches over 100 cM  that appear to be purely artifacts of endogamy, not true cousins. But at least at this level, people I collaborate with who are only 25% Jewish seem to have much more reliable matches. If AncestryDNA says that one of their matches is 117 cM and has some meaningful amount of Jewish ethnicity, that person is almost always going to be a real cousin of theirs. So I use the ethnicity estimate as another piece of the technique: someone who is only part-Jewish, but nonetheless has a promising-looking matching pattern to one of my color-coded groups, is a potentially very strong lead - potentially, an even better lead than someone 100% Jewish with a similar matching pattern. (There are a couple of women in my matches who are probably about 1/8 Jewish, and who I have been itching to hear back from: each of them only has about 7 shared matches with me, but all of those 7 people are known cousins on one side of my family!)
• You can really supercharge the technique if you can convince your cousins who have tested to share your matches with you. Normally, you can't see if that promising-looking lead with a 77 cM match to you also has a 99 cM match to another of your cousins who has tested. But you can if they've let you view their matches! The more sets of cousins' DNA results you can study, the better a look you'll have at who might be descended from a ggg-uncle you didn't know about before.

There is a very informative article that explains the autosomal DNA pitfalls of endogamy, which include Ashkenazim.  The author is Jennifer Mendelsohn.  The title is:  No, You Really Don't Have 7,900 4th Cousins".  There was an earlier reply on this thread from someone who noted that distant cousins could be connected genealogically with a higher incidence of success, if they involved his non-Jewish side.  That was logical.  The non-Jewish population is comparatively heterogenous, such that shared lower levels of cMs are statistically more likely to be from traceable genetic matches.  To avoid the "false positives" of untraceable matches, I use a soft cut off of 150 shared cMs, when dealing with a 100% Jewish person.  However, if a lower-level match has a shared long block of 20 or greater, it is probably not a coincidence.  A shared long block with 20 or more cMs is an indicator of a common ancestor, because it is statistically unlikely to have occurred randomly.  A lot of shared short blocks, which total 150 or below, is a potential indicator of random endogamous DNA that is recirculated among insular, intermarrying populations, such as Ashkenazim.  Instead of being from a single common ancestor, the pattern of many shared autosomal short blocks could be explained by multiple ancestors who existed further back in time, potentially during the patronymic period, which predates the effective generational range of the autosomal test.  (Contrary to the misleading advertising, the autosomal test is good for about 3 generations, 4 if you are lucky.  Beyond that, the signal-to-noise ratio of this test becomes unreliable.  The patronymic period ended in circa 1800 AD, up to 8 generations ago.) 

Ken Domeshek
FTDNA Project Administrator
Damesek, Braverman, Kartorzynski, Sinienski