Context: In various domains such as crisis management, medicine, public safety and mobility, there is a necessity to experiment by using state-of-the-art data driven analytics and decision-making approaches, as well as large quantities of complex data. In the same time, data-driven insights need to be extremely timely, accurate, precise, so that they can be useful to the potential user. Objective: This work deals with the problem of experiment design aiming at the best possible experiment outcomes, such as accuracy and sensitivity of the obtained results. Our goal is to define probabilistic model of data analytics workflow that captures dependencies between datasets, methods and deployment options. The general goal is to create a next generation decision support system that integrates novel research results from the domains of data integration, machine learning, explainable knowledge engineering, and model-driven engineering into a common solution. Method: To achieve our objectives, this work implements a Bayesian network and a Markov decision process for modelling the data analytics workflow, that are used at different stages of the experimental process flow definition. Results: To provide a proof of concept, the testing was performed with a minor Bayesian model built for a dataset with precise constraints, set of four mixed ANOVA methods and set of three deployment options. The results illustrate the practical use of the proposed measure for calculating the conditional probabilities, which allows learning a Bayesian network from the experimenter’s usage. Conclusions: The proposed new solution is appropriate and can be used in decision-making throughout the process flow in complex experiments. Due to its practical utility, this solution will be further researched and improved and in later stages integrated into the ExtremeXP framework.