Social workers are constantly predicting the future. In England and Wales there is a legal duty on them to do so, as the 1989 Children Act requires workers to assess not only whether children "have" suffered significant harm, but also whether they are "likely" to do so. Similarly, in Northern Ireland social workers are required by The Children (Northern Ireland) Order 1995 to do the same. On a more mundane level, social workers are constantly making predictions about whether a parent might use and benefit from a particular service, whether there will be a further incident of domestic violence or even whether a family will be in (or not) for a home visit. Yet predicting the future is hard and doing so with complete accuracy is impossible. Social work is not the only area where prediction is necessary. In the "Good Judgment Project," forecasts made by experts were found, over relatively long timeframes, to be no better than chance. On the other hand, some forecasters were able to outperform not only chance but also highly trained intelligence analysts with access to classified data. Clearly, human judgment is often highly fallible but, in the right conditions, can be incredibly helpful. Might it be possible to improve social work forecasts about the future? This paper considers key issues in theorising prediction in social work, including conceptions of risk, free will and self-determination. It then turns to practical issues, such as the relationship between forecasting and decision-making, and considers some possible research methods and issues associated with them. To illustrate the potential of this approach we describe how we have started to explore the face-validity value of this approach with social workers and how we have measured the accuracy of forecasting in social work.