Copy-pasting my standard explanation whenever this comes up, because I think football gamblers deserve to know what they're into and the vast majority don't.
This isn't how betting works. But don't worry: almost everyone goes through a phase of believing that bookies win by being better at predicting events than the punters. Hell, most people never get out of that phase -- even the most frequent football gamblers I know in real life don't really understand how gambling works. Expert opinion and predictive modelling have nothing to do with the odds. In fact, those things don't matter at all on the bookmakers' side[1]. Bookies have no special insight into anything because they don't predict anything: the prices are set by supply and demand. The more people bet on an outcome, the lower they set the price for that outcome, and vice-versa. It renders odds nothing more than a giant popularity contest.
You could argue that there's a sort of crowdsourced wisdom going on in the results of that process. But then imagine the average punter: he's far from an expert gambler or an expert on the subject on which he's betting. This all leads to prices that very poorly reflect the actual probabilities of the outcomes. Any genuine expertise is drowned out by the sheer volume of casuals, weekend warriors, random-chancers, and so on. (Incidentally, understanding this fact is how "professional" gamblers and true expert analysts play the game: they don't bet on the events they think will occur; they bet solely on the events that they think are significantly more likely to occur than their current price suggests.)
As well as the supply and demand shit, there's a load of money-grubbing maths involved, all of which is now done programmatically by computers, They put a bunch of variables into a formula, including the odds and the amount of money on each outcome, and their proprietary ML-driven algorithms spit out odds recommendations that ensure they make money
no matter what. It's called an overround and it's how the house almost always wins[2] (it's similar to the zero on a standard roulette table: they offer you odds of 1:36 when the actual odds are 1:37, slowly resulting in inevitable losses for everyone over time). Sneaky house edges are why I have no interest in gambling.
[1] They may use predictive models up to a point, to try to refine their overround priorities, train their AI models, and generally get more accurate internal info (e.g. how to set initial prices), but a bettor is
never going head to head with the bookies' predictions.
[2] I say "almost" because unlike a roulette table, real world events, the cumulative patterns of wildly different bettors, and the algorithmic mathematics involved are too complex to
guarantee that they make money, even with a massive overround. When the system fails, you get those glorious stories of bookies losing stacks. Unfortunately, as I'm sure most gamblers know, the cnuts aren't even legally obliged to pay out -- at least not in the UK.