The bolded is not true. If you had done even a cursory googling, you would not assert such. I recommend
this economics senior thesis, which used as its basis ~3700 games covering 3 NBA seasons, applied a reasonable definition and several different measures (ratio over the next N seconds, who scored the next points, etc) and ended up with ~18k usable observations to analyze. Conclusion:
"The most significant of these results, the home-team with the first-half restriction, shows a .21 increase in average ratio for the next ten points, meaning that calling a timeout predicts that the home-team will score 5.47 out of the next ten points as opposed to 5.26 points when a timeout is not called. This result is small, but supports the idea that timeouts can be a marginally effective tool for coaches to use to help their teams win."
A more mathematically complex study from Northwestern is
here. It, too, was unable to reject the null hypothesis that timeouts have no effect on momentum. However, it did find an interesting correlation: the impact that coaches' timeouts have on the teams' short-term performance thereafter depended on coaching experience, and in particular, the more experienced a coach was, the
less of an impact was seen on the team's first few possessions after timeout, relative to their average performance. In other words: experienced coaches are already getting their points across to teams even in the flow of play, and timeouts add little in their ability to convey them. Inexperienced coaches find them to be more valuable. One's interpretation of that conclusion relative to Joe Mazzulla, however, is probably determined by your priors about him, since you can argue it either way.
Meanwhile,
this study on NCAA D-1 data suggests more of a substantial effect from calling timeout, but I am rather unpersuaded by its study design (definition of variables in particular), and in any event there are a lot of other confounding differences between college and the pros.
Regardless, point is, we have done some fairly good evaluations of it, on sufficient sample sizes - and our best guess is that the effect is either zero, or indistinguishable from zero at any meaningful level of statistical significance.