Determining how best to route work is a key element of service system design. Not surprisingly then, many analytical models have identified various optimal routing algorithms for service operations management. However, in many settings, humans make routing decisions dynamically, either because algorithms don’t exist, decision support tools have not been implemented, or existing rules are not enforced. Understanding how individuals make decisions creates the opportunity to identify both positive deviances, as well as suboptimal decision making that can be improved. Therefore, in this paper we first theoretically identify the factors that may impact decision making before empirically examining a large operational data set in a casual restaurant setting to research whether and how hosts deviate from their predefined round-robin rule to seat customers to servers. We find that hosts assign customers earlier than what the round-robin rule suggests to those servers who have low workload, high speed skills but low sales skills, and high familiarity with the hosts. In addition our models reveal that these seating heuristics are suboptimal in our setting and so we suggest an alternative seating heuristic to prioritize servers having high sales ability and estimate a potential sales lift between 2% and 3% through counterfactual analyses. Our research contributes both theoretically and practically as we use empirical methods to show not only how individuals make routing decisions, but also how these decisions can be improved.