I work in so called analytics, using so called artificial intelligence. It's basically the brute force computing of statistical probability, on which a a decision is made. To be more formal, it uses classification, factor analysis, and prediction algorithms...like regressions and neural networks.
It works best in simple tasks. Like playing chess, driving, and most efficient connection flights on planes. It gets much tougher to predict accurately if too many variables need to be weighed in complex tasks...like suggesting what football play to run.
I am familiar with soccer analytics. You can track things like how far a player has run in a game. If he's run > his average early, you can to some degree take advantage of this. If an opponent gives up most goals on the right side with certain players, well, maybe you try to attack more on the right?
As other have said, you really need a human making the final call. A machine just can't adequately fine tune the situation. Maybe tat soccer player on the right gives up the most goals on his team, but also has the most assists due to being an excellent wingback on counterattacks...so do you tempt fate going his lane?
I have to imagine it's the same in football. Humans have a feel, a so called heuristic factor, that machines can't really come up with as well in tight space in a complex task. On that, even though I am not an Xs and Os expert...hell, even I know some silly calls are being made on 4th downs. Probably because coaches and staff haven't been advised as well on analytics.
I could be wrong...but in sports I don't believe the application of analytics is as sophisticated as in, for example, stock trading. You better believe the bots are conservative, with a lot of risk based rules being applied to limit losses. Consider 70% - 90% daily trading asks and bids are by machines...yet they manage to tank entire markets.
They'll blow a play call too...sometimes stupidly.