If your goal is, in part, to foment social discord and political chaos, just breaking in and changing the vote totals is the least effective way and we know what they did was much more effective.
In the former case, you can’t guarantee the outcome but no matter who ultimately wins, you’ve managed to spread social discord and political chaos.
OTH, in the later case, as we do voting in the US, you have to break into countless different voting systems at the county/precinct level without getting caught–with the states and local levels all injecting different people, processes, and systems you have to account for–and manipulate the count in ways that a) are not statistically apparent/aberrant, b) move the correct amount of votes in the correct places, to achieve your desired outcome. And you have to do it on-the-fly during the day or as polls close across the nation and quickly before the “wrong” sets of numbers start getting replicated all over the place. And you can’t mess with things like mail-in ballots the same way or at the same time.
I get that current voting machine and tabulation technology is not very secure. These vulnerabilities do not add up to a credible way to reliably flip a national election. (Elections of smaller scope, even statewide in some states with more uniform statewide voting systems, seem like they would be much more vulnerable.) To drive a national election to a specific outcome, social engineering seems much more likely–indeed I’d say was proven in 2016–to be effective.