Without a process you're using correlation implies causation. Which can lead to a lot of wrong turns. Especially when that correlation is based on anecdotal n

Without a process you're using correlation implies causation. Which can lead to a lot of wrong turns. Especially when that correlation is based on anecdotal non repeatable massively multi variate data like you said. Could be true, could be a dead end, don't put as much weight as you did on it.