Nowadays, Energy Management Systems (EMS) are accessible for homes and buildings to optimize energy consumption especially when solar panels and batteries are installed. The intelligence of existing systems is often based on environmental or exogenous information like the weather, energy prices, and endogenous information like user consumption behavior and activity. The solutions aim to adapt a consumption profile to the produced energy in order to reduce costs. In the case of a perfect prediction of all variables, system performance can be controlled. In this article, we study the impact of generation prediction error on the daily energy cost. For this, we consider the energy management system as a black-box and we simulate multiple scenarios with different prediction errors using the quasi-random Monte Carlo method. We observe the global sensitivity of the system by measuring the Sobol indices in order to identify errors that impact more the daily energy cost. The analyses are based on French consumption data and on irradiance data for Carpentras, France. Results show that findings are aligned with battery charge and discharge strategies.