The revenue impact of a feature is based on two things; 1) the feature's isolated individual effect and 2) any synergies the feature might have with other features in that game (e.g. guild mechanics + co-op PvE = more than the individual effects' sum).
​
The individual effect of a feature is based on our mathematical model, which takes into account for example the game type (e.g. puzzle game) and how much that feature differentiates best performing games (top grossing-wise)) from no-so-well-performing games. This way we're not only looking at what best performing games are doing, but rather that what features and feature combinations actually make difference when it comes to top grossing success in certain game types.
​
The synergy effects are based on our machine learning model which is able to find especially powerful (or bad) combinations when it comes to driving games' top grossing performance. As we have the same 200+ features tracked from thousands of games, there are countless of possible combinations and that's where machine learning comes handy.
​
So as a rule of thumb and as a rough simplification, the more a feature differentiates best performing games from the rest, the higher the effect is on revenue. Of course synergies, game type and lots of other things are taken into consideration, as games are quite complex beings, but I hope this gives you a better idea of the calculation process