Three CMEs which erupted on 2017 Sep 4 and 6 underwent mutual interaction before reaching Earth on Sep 6-9, where it gave rise to a complex and unexpectedly geoeffective structure as detected by WIND at L1. The spacecraft first observed an interplanetary (IP) shock on Sep 6 followed by a turbulent sheath. The leg of the CME flux rope is detected on Sep 7, in which clear signatures of a shock-in-a-cloud can be distinguished, coming from the third CME which propagated into the preceding flux rope. We model the source of this complex ejecta with WSA-ENLIL+Cone and EUHFORIA to assess and compare the overall performance for interacting CMEs as opposed to single CME events. We find that following the conventional algorithm for determination of input parameters give large deviation in the event prediction at L1. The overestimated density of the IP shock 1 is due to the way of implementation of the magnetogram in WSA model. Excluding the slow CME from the input leads to even larger deviation. The prediction of IP shock 1 drastically improves by introducing of a customized density enhancement factor (dcld) based on coronagraph image observations. This novel approach, is simple and accessible, and could be applied to improve the forecast for fast, undisturbed CMEs. The deviation in the prediction of IP shock 2 comes from its interaction with the low proton temperature environment of the preceding magnetic cloud, giving rise to an expansion of the shock front. Additionally, the properties of the background solar wind plasma have been preconditioned by passage of the previous IP shock. This was confirmed from the kilometric type II radio burst emission following the eruption of the third CME. The propagation profile of this CME implies an almost non-existent deceleration in the interplanetary medium, in contrast to the expected CME deceleration due to interaction with the background plasma. In summary, this study presents clear indications that magnetic interaction must be taken into account to reliably forecast multiple CME events. Preconditioning of previous CMEs must also be considered in more depth, and ultimately requires a realistic, time-dependent model of the ambient solar wind which responds well to propagating shock waves. Models in space physics presents us with the perfect tools for understanding not only the physical processes that the simplified models can predict, but perhaps more importantly, help us begin to understand what the models fail to predict.