Background, A/B checking is a regular measure in many marketing procedures for ecommerce companies. Through well-designed A/B research, advertisers can gain insight about when and how marketing efforts can be maximized and active promotions driven. In practical terms, standard A/B experimentation makes less money relative to more advanced machine learning methods. Purposes, in order to examine the current A/B testing state, identify some popular machine learning algorithms (multi-arm bandits) which are used to optimize A/B testing, and then explain the output in some standard marketing cases of these algorithms. Methodology, In this study, the state of A/B testing have been addressed, some typical A/B learning algorithms (Multi-Arms Bandits) like Thompson Sampling, Epsilon Greedy and UCB-1 will be implemented and compared used to optimize A/B testing are described and comparable. As a result, UCB-1 and Thompson Sampling, be an exceptional winner to optimize payouts in this situation. Because it showed more effective results, without losing experimentation and statistical variations, to maximize total payouts. Based on its accuracy and strong tolerance to noise on the results, UCB-1 is the right option for MAB for a low base conversion, a limited impact size scenario.