Optimizing the use and maintenance of battery dependent systems is of great economic and ecological importance. In practice, however, batteries are often replaced too early to avoid downtime at all costs. As a result, valuable capacity is thrown away, leading to unnecessarily high costs and ecological impacts. To address this challenge, this work presents a comprehensive end-to-end workflow for predicting battery discharge behavior using sparse multivariate time series data. The proposed methodology encompasses five key steps: 1) Data Acquisition, 2) Augmentation, 3) Preprocessing, 4) Feature Engineering, and 5) Modeling. One of our key contributions is to enrich the limited available training data by applying augmentation techniques. In extensive experiments, including 270 evaluations across Survival Analysis and Iterative Regression, we demonstrate the effectiveness of our method. In particular, we show that incorporating augmented data improves the Integrated Brier Score by up to 50.16%. Moreover, we propose a novel, use-case-centric metric called Mean Divergence Time which estimates how long time series predictions remain reliable within a defined tolerance. Our results show that an accurate modeling of battery discharge curves in our use case is possible for an average of 117.76 days. This provides sufficient time to optimize the planning of maintenance intervals. Thereby, our work contributes to the modeling of battery lifetimes with sparse data and sets the stage for future research on the use of augmentation techniques to increase efficiency and resource optimization.