There is concern that rather than critically deliberating specific circumstances, auditors focus on selecting and documenting defensible audit positions. Currently, subordinate auditors perform tasks mindful that they will be accountable for their work both inside (e.g., partners) and outside (e.g., PCAOB) the firm and adopt “implementation intentions” based on previous review experiences to guide their performance. In the context of fraud‐detection planning, we consider an alternative approach in which subordinate auditors work under contingent reward agreements under which they will be compensated for effective fraud‐detection plans. Lacking an anticipated course of action, they invoke a “deliberative mindset” in order to create a task strategy. In an experiment, auditors completed a fraud‐detection planning task under contingent rewards, accountability, or anonymity. We find that auditors operating under contingent rewards used deliberative mindsets. They were better able to identify potential fraud, select more effective procedures, and plan more hours for effective procedures. Auditors under accountability completed the planning task based on implementation intentions. They focused on broadly increasing audit hours across procedures, including allocating significantly more hours to less effective procedures. Mediation analysis shows that improved planning performance resulted from the use of deliberative mindsets and not implementation intentions.

Although lower-level auditors increasingly carry out mandatory analytical procedures (APs) in audits, they do not perform as well as partners and managers. In order to improve performance in APs by lower-level auditors, we investigated tasks requiring creativity, where training in metacognition—consciously thinking about one's thought process—improves task performance. As a result, we train lower-level auditors to use a sequential thought process comprised of two metacognitive skills: divergent thinking, where they generate explanations for unusual evidence, followed by convergent thinking, where they evaluate explanations generated and eliminate those judged infeasible. To test the efficacy of our training, we conducted an experiment with three conditions: both divergent and convergent thinking, divergent thinking only, and a control. We found that training auditors in only divergent thinking increases both the number and quality of explanations generated for an unusual situation. However, the combination of divergent and convergent thinking training leads to improved explanation generation over divergent thinking alone and, more importantly, leads to a greater likelihood of generating and choosing the correct explanation.