The evaluation is only helpful for detecting mistakes and blunders!

If you actually want to learn the optimal moves for your opening, or see what would have been the "best move", you're out of luck. Here's an example:

1. e4 e5 2. Nc3 Nf6 3. Nf3 

I played Bc5 in the game. That's an inaccuracy, so let's see what would be the best move: "Bb4 is best -0.02".

Let's go back and see where to go from there, for future reference. I play Bb4, expecting '-0.02' but that's the problem. Now the engine says "Bb4 is good -0.41, but Nc6 is best -0.30". 

This was the "best move" before, and now it's just good. If I do play this "best move" the evaluation of it says "Nc6 is a book move -0.43, but Bc5 is best -0.47". 

Right now it's suggesting a move that, according to the number, is worse for me as the best move. I can forgive that because it's quite close. However keep in mind that I've come full circle now. It keeps suggesting a different best move every time. It turns out there is no best move, because when you play it you get "okay" or "inaccuracy". 

This is not just something that happens in the opening. I've seen it mid game and end game too. There's just no move the engine likes. It dislikes what you played, but when you evaluate and try out the suggested alternative, the engine doesn't like that either!

When you actually follow the engine line by another move, you get in a worse position. If you keep playing the best move for yourself and the opponent, your position keeps getting worse and worse.  

I find that the only usable part of the evaluation are for finding blunders or missed wins, and for getting a general evaluation of the position. Anything other than these things, the engine is just useless for. Feel free to share your thoughts, or parhaps give me some advice if you have any.