Human decision making in the newsvendor context has been analyzed intensively in laboratory experiments, where various decision biases have been identified. However, it is unclear whether the biases also exist in practice. We analyze the ordering decisions of a manufacturer who faces a multiproduct newsvendor problem with an aggregate service-level constraint. We find that the manufacturer broadly exhibits the same biases as subjects in the laboratory and is prone to another bias that has not been identified before, that is, group aggregation. The bias can be attributed to the multi-product problem of the manufacturer, and refers to the observation that the service levels are not optimized for individual products, but rather for product groups. Our data allow us to analyze the performance of a manufacturer in detail and we find that target service levels are achieved effectively, but not efficiently. We provide rationales for the manufacturer's ordering behavior, discuss managerial implications, and quantify the financial benefits of debiasing ordering decisions.