The climate is changing; this will impact species, but how? A multitude of alternative models from scientists paralyses decision makers and prevents action on climate change. Using artificial intelligence techniques, we developed a way of 'learning by doing', that finds the most likely model over time and recommends the most effective management action. For a globally significant population of migratory birds threatened by sea level rise, our approach protects 25,000 more birds per year than current best practice. Uncertainty should not impede immediate action on climate change: we demonstrate how managers can act confidently while scientists improve predictions.