Interpretive Structural Modelling (ISM) is widely employed in production research to study the complex interaction among various factors or elements which define a complex production or supply chain problem. It transforms the poorly articulated mental model of the problem into a visible well-defined relational model using an element-relationship-matrix. Building ISM involves primarily pairwise comparison of factors in rotation i.e. each factor is compared with all remaining factors as input. In general, these relations among the compared pairs are defined in binary levels i.e. the relations are defined in terms of “yes/no”; hence, the interactions are treated equally for all levels of interaction magnitude. Consequently, the interpretation of the results does not capture the intensity of interrelation, which limits the exploitation of the relational model for concrete production/supply chain decision-making. This paper introduces a data-driven algorithm to convert a multi-level pairwise comparison into bi-level groups i.e. groups with weak and strong relations, to incorporate and account for non-binary relations. The bi-level groups are created based on a threshold point in multi-level input that simultaneously maximizes the inter-group variance whereas minimizes the intra-group variance. The application of the proposed approach is demonstrated in context to small-series textile/apparel supply network configuration, in order to show its practical significance in strategic decision-making.