We propose a novel methodology for analysing change propagation in software using the domain-level behavioural model of a system. We hypothesize that change propagation analysis is feasible based purely on the information visible and understandable to domain experts, trading some accuracy for productivity. Such a method is independent of formal architectural representations and may be practical for applications with heterogeneous subsystems, or missing or undocumented source code. In this paper we introduce the first phase of the methodology: creating and evaluating a connection graph of conceptual relationships between user interface components. We provide results of case studies on two web-based systems which illustrate how our methodology can be applied, and how discovered conceptual relationships match the architectural dependencies.