ABSTRACT Unbiased and precise mass calibration of galaxy clusters is crucial to fully exploit galaxy clusters as cosmological probes. Stacking of weak lensing (WL) signal allows us to measure observable–mass relations down to less massive haloes without extrapolation. We propose a Bayesian inference method to constrain the intrinsic scatter of the mass proxy in stacked analyses. The scatter of the stacked data is rescaled with respect to the individual scatter based on the number of binned clusters. We apply this method to the galaxy clusters detected with the AMICO (Adaptive Matched Identifier of Clustered Objects) algorithm in the third data release of the Kilo-Degree Survey. The results confirm the optical richness as a low-scatter mass proxy. Based on the optical richness and the calibrated WL mass–richness relation, mass of individual objects down to $∼\! 10^{13}\, \mathrm{M}_⊙$ can be estimated with a precision of ∼20 per cent.