Detecting and measuring the masses of Earth-like planets in the presence of stellar signals is the main challenge when using the radial-velocity (RV) technique. Even in the PLATO era where the satellite will provide the period of Earth-like planetary candidates, measuring precisely their mass, which is critical to 1) confirm those candidates, 2) constrain further planetary composition and thus planetary formation and 3) constrain further planetary atmospheres, will be extremely challenging. Critical to a better understanding of RV variations induced by stellar signals and finding correction techniques is RV data with a sampling and SNR sufficient to probe stellar signals ranging from minutes to years. To address this challenge, we can use the unprecedented data from the solar telescope that feed sunlight into HARPS-N, which allows us to obtain Sun-as-a-star RVs at a sub-m/s precision. In this talk, I will discuss how to reduce properly the HARPS-N solar data to reach a precision of about 50 cm/s on the short and long-term. This implies optimizing the wavelength solution recipe, carefully selecting the most stable thorium lines, but also compensating for the ageing of thorium-argon lamps inducing a drift of thorium lines with time. I will show how those optimizations improve the quality of the data, and therefore will advise any team working in extremely precise RV to perform similar upgrades. The obtained solar data, published last October, have already been used in several studies that demonstrate that analyzing the HARPS-N solar spectral (or cross-correlation functions) time-series using machine learning algorithms can mitigate stellar signals down to a level where Earth-like planets in the habitable zone could be detected (30 cm/s in semi-amplitude, signal three times larger than Earth).