Product recommendations are generally understood as data-driven – however, we argue that knowledge-driven management decisions may also play a role, especially in the cold start problem, which has been tackled with various degrees of success through a number of approaches. We hereby advocate an approach that captures managerial priorities in the act of recommendation building – i.e., the proposal is to complement the traditional customer-centric view (affected by uncertainty) with a machine-readable business-centric view. For this purpose, the paper reports on an engineered method for the “digital nudging” of recommendations - it starts by capturing a manager's priorities with diagrammatic means, which are further exposed as a Knowledge Graph to a recommender built on a modified version of the Onicescu method taking into consideration a business “utility” concept to influence decision-making. The research follows the Design Science methodology, resulting in a “method” artifact that tackles the cold start with the help of a (by-design) recommendation nudging mechanism. In terms of method engineering, the proposal orchestrates its ingredients into a coherent method with the help of (a) Agile Modeling Method Engineering, to setup up a diagrammatic tool for prioritization rules, (b) the Resource Description Framework, to capture the diagrammatic rules in knowledge graph form, and (c) the Onicescu multi-criteria decision method with modifications based on Zipf's Law. The evaluation was based on surveys with potential stakeholders, for the different steps of the method. The implications are that the notion of “digital nudging” can take a knowledge-driven form, engineered as an artifact that bridges the decision-makers' priorities (captured by diagrammatic means) with the customer-facing output (recommendations), instead of relying solely on the accumulated history of transactional data. This interpretation of digital nudging may be extended towards other “digital choice environments” where contextual decisions are called to influence the computational output.
