More and more people are choosing hospitals as their first place of admission when they fall ill. This trend has continued to increase over the years, resulting in overcrowded hospitals. Many solutions to hospital overcrowding have been proposed. These include trying to discharge patients as early as possible and reduce their length of stay (LoS), which is achievable through precise planning without compromising the quality of treatment. In this paper, we simplify planning by automatically predicting hospital occupancy both in the emergency room and later, when patients stay on the ward. This approach relieves hospital staff of planning tasks, allowing them more time to care for patients. The prediction is done by aggregating the predicted LoS with an estimate of how many patients will arrive in the future and their expected LoS to determine occupancy. We demonstrate how the accuracy of LoS predictions affects the accuracy of occupancy predictions. We evaluate our approach using the anonymized MIMIC-IV EHR (electronic health record) database and successfully apply it to a real-world scenario at another hospital in Germany.