Across the UK, child protection social workers are routinely called upon to assess the likelihood of future significant harm to children. Yet making consistently accurate judgements about what may or may not happen in future can be a difficult task. In a previous study, we tested social workers' abilities (n = 283) to forecast the likelihood of different actions, events and outcomes following real-life referrals to social services. Aggerate group performance was only 6% better than you would expect by chance. As a result, we wondered whether social workers could make more accurate forecasts in relation to families they know well. In this paper, we report the results of an in-depth case study, involving two social workers and five families. For eight-months, the social workers generated their own forecasting questions and provided estimates about the likelihood of different actions, events, and outcomes. One of the social workers, with more experience, made forecasts that were on average 12% more accurate than you would expect by chance. The other social worker, with less experience, made forecasts that were 6% less accurate than chance. These findings suggest that simply having more information about the family may not make a consistent difference to the accuracy of social work judgements. What we do not know is the extent to which these findings might be replicated with a larger sample, or the nature of any potential relationship between more accurate judgements and better decision-making within the complex ecology of social work.