Abstract. Realistic cloud droplet number concentrations (Nd) are critical for simulating the Earth's radiative budget, yet remain challenging to represent in global models. One reason is that aerosol optical depth (AOD), commonly used for aerosol validation and assimilation, only weakly constrains aerosol number concentrations relevant for cloud formation. We introduce a representation of aerosol-cloud interactions (ACI) into the ECMWF Integrated Forecasting System (IFS), restricted to aerosol activation and the first indirect effect, and use it as a diagnostic tool to relate aerosol properties to cloud observations in the Copernicus Atmosphere Monitoring System (CAMS). Using 18 years of MODIS Nd retrievals, we infer effective aerosol size distribution parameters and evaluate results using independent observations of AOD, Angstrom exponent, size spectra, and top-of-atmosphere shortwave fluxes. After optimisation, CAMS aerosols produce more realistic large-scale Nd patterns, but exhibit regional biases: overestimation over sub-Saharan Africa and underestimation at high latitudes. The African bias is consistent with carbonaceous aerosol emissions from wildfires and can be partially mitigated through optimisation, while the high-latitude bias is likely linked to excessive aerosol scavenging in mixed-phase clouds and cannot be resolved by size adjustments alone. We show that introducing phase-dependent aerosol scavenging substantially improves all-sky shortwave fluxes over the Southern Ocean compared to the ACI implementation in the standard CAMS system. This highlights a direct link between radiative biases and aerosol wet removal processes. Representing ACI in a weather model enables Nd to provide additional constraints on aerosol numbers and related processes globally, revealing limitations in how ageing, vertical transport and wet removal are represented, as these processes are only weakly constrained by standard AOD evaluation.