There have been few articles describing business optimization using recommendation engine approaches for slot games in online gambling, and there is not much information available about which approach works best in that scenario. Along with an existing online gambling software platform, we want to offer our customers the necessary recommendation logic to optimize their business. We create a game recommender system using cloud technologies and data about players and their playing behavior. We describe how we can integrate it into the current software architecture used in the company. We then test different approaches, evaluate different algorithms using historical data, and explore techniques of assigning algorithms to different players dynamically. After running the experiments, we find that the Alternating Least Squares (ALS) algorithm produces the best recommendations for our scenario. We test the algorithms using the Leave-One-Game-Out (LOGO) testing approach. ALS successfully predicts back 63.4 percent of games that we remove from the training set. We also conclude that pre-clustering players does not help much with the precision of recommendations. In the case of ALS' LOGO test, it contributes nothing. Finally, we show that it would be good to use the Multi-Armed Bandit (MAB) Softmax algorithm in a production test scenario to avoid losing too much precision.