Realistic number concentrations of aerosol cloud condensation nuclei (CCN) are critical to simulate the Earth&rsquo;s atmosphere radiative budget. However, CCN concentrations are very uncertain, which critically limits our capacity to precisely predict climate change. This is partly because aerosol optical depth (AOD) observations, traditionally used for validation of and assimilation into aerosol models, weakly constrain number concentrations. We introduce aerosol-cloud interactions (ACI) in the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) to simulate the activation of aerosols into cloud droplets driven by aerosol concentrations from the Copernicus Atmosphere Monitoring System (CAMS), using offline computations from a parcel model. Then, we use an 18-year-long series of MODIS cloud droplet number concentrations (<em>N_d</em>) retrievals to tune aerosol size distributions and simulated number concentrations. Finally, we validate the system using AOD observations, all-sky broadband SW fluxes, Angstrom exponent and size spectra from satellite and AERONET. We found that CAMS aerosols allows simulating overall realistic <em>N_d</em> values, but these are systematically overestimated over most of sub-Saharan Africa and underestimated at latitudes higher than 60&deg; N and 45&deg; S. While the first bias hints at issues with carbonaceous aerosols from African wildfires, the high-latitude bias originates from too efficient aerosol scavenging in mixed-phase clouds. We finally show that addressing this last issue dramatically improves simulated clouds over the Southern Ocean. This study showcases that representing ACI in a weather model is a powerful diagnostic tool to improve aerosol representations and processes, potentially informing also applications in climate models.