Most cheaters get banned because they use cheats easily found online. Anti cheat makers go around buying access to all the new cheats so they can reverse engineer them to figure out how they work then they change their anti cheat to combat it.
In theory someone who's good at software engineering can build their own cheat and as long as they don't share it, it can go undetected for years
To be honest, a lot of the larger game studios are banning based more on player data, as opposed to trying to detect bots or wallhacks, because you can't really win by doing that in the long run.
For example, if you start playing a game right now, say CS:GO - your performance will fall to within an expected range, by performance we're talking every element of your play, how you move, how many headshots you can manage, the skill difference between you and the people you kill - etc. You can generate a number or value based on the aggregate of all of those statistics, and assign a skill level to it.
If you play for a few days at that level, then download a cheat - and your performance instantly exceeds that of a top 500 player in some metrics, it's obvious you're cheating from the data. You don't need a screenshot or admin review, the data proves you're playing at an impossible level - it's that big jump in skill level that gives it away, because people just don't play like that normally.
When I worked at Riot it was obvious in League of Legends, most players who cheat - download the game, play for around 10-15 hours - get owned and download a cheat, then their skill level skyrockets to impossible levels, they get banned. They download the game again on a new account, do the same thing again and again. In the end Riot stopped trying to ban every new cheat account, and simply had the matchmaker put them all into the same game, so we had thousands of servers running which were just full of cheaters playing against cheaters, whilst keeping clean players on separate servers only being match-made against other known players of good standing.