Tracking down the root cause and impact scope of end-to-end network service anomalies is challenging. We present a generic approach to fully automate this process of scoping the service impact of a network event within a fine-grained multidimensional time series dataset. Our approach begins with a metric definition which models how well an anomaly fits across relevant dimensions. The metric guides AutoFocus, a greedy algorithm we designed to significantly reduce the search space. We evaluate AutoFocus on a real-world multidimensional datasets from a large network operator, using both synthetic event generation as well as case studies. Our evaluation shows that our approach results in accurate trouble isolation, thus helping to automate the process for network operators.