The video is quite educational, but they wrong on Ancestry's Timber algorithm, because it doesn't work well. It reduces and completely deletes matches which are real and vice versa. It sometimes works just fine, but due to reducing or even sorting out real, valuable matches, it causes more problems than it solves. Similarly, FTDNA and 23andMe have a better representation of the matches, even if you have to sort it out yourself, what's real and what's not.
Timber is just great in theory, not in practise.
She is also much too negative on small segments, since even though many are not real, many others are and some can point to actual, provable genealogical relatedness. I was able to verify segments as small as 3 cM in specific instances, yet alone 7-10 cM sized segments.
Of course, if you want to go for very high matches (90 cM and higher) only and have lots of matches, you don't have to care. But if you want to trace more distant relationships, then its not recommendable to rely on Timber and disregard small segments (< 20 cm).
However, for actual endogamous people, small segments can be oftentimes useless, yet even that depends and can't generalised for all cases.