We propose a new method to model the number of events that occurred in the past, but which are not yet registered due to an observation delay. Our approach provides an elegant and flexible framework for modeling the observation delay subject to calendar day covariates by introducing the concept of observation exposure.