Many problems in business decision making can be modeled as mathematical optimization problems, like mixed-integer programming problems; or satisfiability problems. However, the solvability often depends on the structure of the encoded problem, and from a theoretical computer science perspective, both mixed-integer programming problems and satisfiability problems belong to a family of hard problems. Moreover, with the advent of big data, the instances, which are needed to be solved are becoming lager and larger. Thus, to tackle the issues imposed by theoretical hardness and ever-increasing problem-sizes, further improvements in the available solution algorithms are needed. In the OPTIM-AI project, we investigate how the existing solution algorithms can be hybridized with approaches from artificial intelligence in order to deal with these challenges. In this talk we give a brief overview of two interesting developments, namely i) the use of deep learning with graph convolutional networks for combinatorial optimization problems on graphs, ii) the use of decision diagrams for solving optimization problems.