Identity and access management (IAM) solution is responsible for access (de-)provisioning of users in enterprise environment. The digital transformation has expanded and complexified the IAM problem, therefore the traditional manual approval process is tedious, expensive and not scalable. We designed and developed a recommendation system to automate IAM problem in an environment with more than 70000 employees and 80000 roles. Our recommendation system combines role-users similarity in form of similarity on a graph and a deep learning model.