The inherent uncertainties in numerical reservoir simulation can lead to models with significant differences to observed dynamic data. History matching reduces these differences but often neglects the geological consistency of the models, compromising production forecasting reliability. To address this issue, this work proposes a geological modelling workflow integrated within a probabilistic, multi-objective history-matching workflow, using the concept of pilot points. The pilot-point method is a geostatistical parameterization technique that calibrates a pre-correlated field, generated from measured values, and a set of additional synthetic data at unmeasured locations in the reservoir, referred to as pilot points. In this study, the synthetic data correspond to synthetic wells; henceforth referred to as pilot wells. The methodology is applied to a real dataset, the Norne Field benchmark case. The flexibility of the pilot-well method is the principal advantage, while a key challenge is to optimize the pilot-well configuration. The configuration includes production data, the preferred fluid-flow paths and the geological framework. The flexibility of the method is demonstrated in the two case studies presented here: generating specific sedimentary features (G-segment) and finding the best location for the cemented stringers responsible for the fluid behaviour (C-segment).