The behavior of complex systems is intrinsically difficult to model and to predict. Agri-food chains can be considered as such. The problem considered in this paper comes from the commitment to anticipate the impacts of the implementation of innovations in such highly complex agri-food systems. The paper focuses on comparative methodological issues, when seeking to anticipate the possible evolutions of the system. Indeed, this article proposes adaptations in the classic "scenario method" because of constraints of remote working generalized during the pandemic, and discusses possible biases induced by these adaptations in the results obtained. The methodological and organizational differences are described, and show that the remote constraints do not prevent from delivering some "key variables" of the system. The adapted method is illustrated in a case study in the pork supply chain. Nevertheless, the face-to-face collaborative sessions generating a consensus among players in the classic method can not be replaced in the remote context. As a consequence, it is likely that some key variables that would have been selected thanks to consensus in the classical method are let aside in the adapted method, because the number of prospects quoting them spontaneously in individual interviews is not large enough. The following consequence is that the scenarios that would have been generated thanks to the various values of these likely key variables, are not taken into account. It is thus likely that less scenarios are depicted by the adapted method than by the classic one.