Demand forecast accuracy improves management earnings forecasts and is critical to organizational planning and coordination. Indeed, surveys indicate that CFOs name forecast error as their top internal concern and identify demand forecasting as one of their top organizational priorities. In this study, we examine organizational learning effects over time in forecast revisions. We first document that year-over-year experience in product level demand forecasting improves forecast revision quality; that is, we show forecast revision “learning by doing.” We further examine whether the disaggregation of a demand forecast into separate forecasts for each source of demand facilitates learning by doing. Exploiting proprietary data from a multinational manufacturing organization we show that disaggregation leads to higher quality forecast revisions that result in improved forecast accuracy earlier in the forecast cycle. Our study provides new insight into how organizations can begin to address the vexing problem of demand forecast error.