One important way referenced in the article is to evaluate failures as well as successes for the likelihood of a trait: for example Jim Collins describes certain features that successful companies have, but does not evaluate the pool of losers to see if they also featured the same trait.

Another example is when evaluating various corporate traits against stock market performances, many people will choose a list of companies that exist today and then study their history to identify correlations of various traits. This omits those that no longer exist today, and it is a common problem among, for example, beginning hedge fund equity analysts. Instead, one should study the cohort from a random sample as existed at the beginning of the study, not at the end, and then study their performances from that point forward.

Agreed with mediaman here.

Select both successes AND failures. Where they seem similar, those things may not teach us much. Where there is a pattern shown ONLY IN ONE SIDE (more or less) there might be a lesson.

I have a fear that there's isn't a lot you can do to guarantee success, but I admit this is just my feeling, not based on any facts or stats.