With the recent boom in machine learning (ML), people are interacting with a growing number of ML-advised products in many aspects of their everyday life. However, the underlying logic in the decision-making process of ML systems generally lacks transparency. This becomes an issue when the ultimate decision is made by the end user and the output of the ML tool disagrees with the user’s belief system. Such disagreements, combined with incomprehensible ML reasoning, cause difficulties in human-AI interaction (HAII). One domain in which such predicaments often occur is the choice coordination of mixed-marketing plans (MMPs). ML methods like probabilistic graphical models (PGMs) can be used to optimize a MMP’s effect on the key performance indicator (KPI) of a company, but the optimized MMPs appear synthetic to marketing employees, who are therefore reluctant to adopt these recommendations. This thesis aims to mitigate such gaps between ML and user in the context of marketing planning and HAII. The contributions of this thesis are: a) An approach to combine PGMs and neural networks (NNs) through approximation. This combination enables rapid feedback from the AI-recommender system to the user. By extension, this affords additional human-AI feedback loops before reaching user fatigue. In turn, this provides the users with a better understanding of the AI behavior. b) The introduction of the NN-based game iNNk to be used as a case study for HAII. This study gives empirical insights on how humans and NNs perceive and classify the same data differently. These differences cause game-breaking player strategies to emerge. c) A simple method to ease identified limitations of the NN in iNNk. These limitations were discovered by observing HAII with non-expert users playing the game. d) A stepping stone toward ultimately combining the previous insights to augment the AI-recommender system used for marketing planning. This stepping stone is a method to modify the HAII flow in marketing planning by enforcing diversity. This diversification opens new possibilities, as it allows marketing employees to develop an improved mental model of the recommender system. In addition, it allows for better adaptation of user preferences in complex, high-dimensional optimization tasks.