Software-as-a-Service (SaaS) provide services based on various user subscription options that have fluctuating resource demands. To maximise profit, income must be maximised, and expenses must be minimised. SaaS owners can optimise their profit by aligning subscription prices with their resource demand. However, because different subscription configurations involve different features, monitoring their resource demand can be challenging. Our proposal aims to estimate the resource usage incurred by users and use this information to determine the demand of each subscription configuration. To do this, we propose a machine learning approach that determines the resource usage of the system at a given time, while having access to the volume of different subscription users, thus paving a path to subscription resource usage. The model is first trained with data monitored from simulated user activity in a preproduction environment. Then it is updated with real user interactions obtained from minimally invasive monitoring in the production environment. The effectiveness of the solution is gauged by the model’s error metrics and an evaluation of the training data. This solution enables us to assess the resource demand of each subscription configuration based on the resource usage estimate.