In this report of an RMetS meeting in February, Nick Dunstone and Doug Collins from the Hadley Centre cover a few points that take us beyond the article above. Full details of the presentations can be found at: http://www.rmets.org/events/meeting/archive.php Trevor Maynard of Lloyd&'s gave an overview of insurance industry needs. Some of the largest insurance losses are caused by climate-related events, with record claims coming from hurricane Katrina in 2005. Catastrophe (CAT) models are key tools for assessing risk and would benefit from skilful climate predictions to ensure stability of insurance premiums. Realistic disaster scenarios are also being developed, many of which focus on weather-related risks. The insurance industry is keen to encourage further research and collaboration to improve risk management, and Lloyd&'s have recently introduced an annual Science of Risk Prize (http://www.lloyds.com/scienceofrisk). Adam Scaife highlighted the successful prediction that 2010 would more likely than not be the warmest year on record for global surface temperature, and described recent work showing skilful multi-year predictions of Atlantic hurricane frequency and ongoing research into the prediction of extremes. Adam and Ed Hawkins both highlighted the importance of the high-latitude North Atlantic region, shown both in modelling studies and statistical predictions. This is consistent with the predictability of the Atlantic meridional overturning circulation, identified using multi-model reanalyses and prediction experiments. A strong message throughout the meeting was the need for careful analysis of accuracy and reliability in order to build confidence in decadal predictions. Tim Palmer stressed that ‘reliability’ has an exact meaning: if an 80% probability is forecast for an event, then over a long series of forecasts the event should occur 80% of the time. Andy Challinor outlined the End-to-end Quan-tification of Uncertainty for Impacts Prediction (EQUIP, http://www.equip.leeds.ac.uk) project, which aims to assess and improve the treatment of uncertainty and develop risk-based prediction for decision making. Several speakers discussed potential improvements in decadal forecasts. Ed Hawkins suggested that optimal perturbations might improve forecast reliability. Len Shaffrey outlined some of the potential benefits of higher resolution seen in the HiGEM model for improving the mean state in the equatorial Pacific and for storm development and tracking. Adam Scaife showed that higher horizontal resolution improves the simulation of blocking, and a better representation of the stratosphere improves the simulation of important teleconnections. Such model developments are, of course, more expensive, and Tim Palmer made a strong case for improved computer power for decadal prediction: a number one computer for a number one problem.