I'm working on developing an American Football simulation engine. Right now, everything is still being designed on paper.  So far, I can see easily how I can create a team, and create individual players with their ratings. I can even see creating the stadiums and weather conditions to affect how they play.  My next big hurdle is figuring out how to simulate playing a game. 

I have been looking around for a tutorial but can't find one for this type of thing.

The only way I can think of to start this, is to create a coach class which would make all of the in game decisions based on how the game is going. If I do that, I would pretty much need to create plays, and a playbook.
This seems like an awful lot to generate some numbers. I feel like maybe I am over thinking this?  I mean games like Tecmo Super Bowl III for the SNES had decent season simulation so I feel like there must be an easier way?

Any help pointing me in the right direction would be greatly appreciated!

Thanks.

You should read some about AI design... in a few words, you need to firmly define what are the individual actions--namely "states"--a player would do, the less actions the more reactive would the AI look like at a first glance. Actions are localized through some UID (unique identifier). A second important bunch of definitions is the so-called "inputs" each individual player would receive--actually in a FIFO order--those are e.g. other team is holding the ball, an opponent is very close to the player, etc. but in a really "live" fashion so that you think it is in the real game.   Those "inputs" also have their UID.

With those 2 sets of UID, try to find out a game strategy/tactic. THE BEST WAY IS TO OUTPUT A LOGICAL SCHEME that tells when the player changes from state A to state B and how it goes from B to A back, thus depending on inputs (notice that the fewer inputs, the lesser smart the player can react).
 
From such a scheme, there are various wayd of implementing it... I do prefer a sequential digital design which offers best performance result over complexity of the schemes (like robots would be designed).