It has been well recognised since the 1980s that translocations should be monitored so we can learn from their outcomes. However, the nature of this learning usually remains unclear. While there have been numerous attempts to summarise translocation outcomes, this generally results in summaries of success rates that have little or no predictive value for future decisions. We argue that learning is most effective when done in a formal adaptive management framework where predictions are progressiveness improved through Bayesian updating. We further argue that translocation programmes are inherently Bayesian because they should involve: 1) collating the data available when a translocation is proposed to make PRIOR predictions to guide whether it should to proceed; 2) collecting NEW DATA after release if a translocation proceeds; and 3) using the data to update the original model, allowing POSTERIOR predictions that can be used guide ongoing management of the population and to improve prior predictions for the next proposal. We illustrate how this process has been used to guide reintroductions of the toutouwai (North Island robin, Petroica longipes) to predator-managed areas in Aotearoa New Zealand. We discuss the challenges involved in extending this formal updating process to multi-species programmes.