The paper quantifies the precision gains in forecasting CPI headline inflation by grouping the set of CPI items as homogeneous groups and by forecasting the inflation rates of those CPI sub-aggregates before aggregating them. We obtain two sets of homogeneous groups by applying a clustering method to both the whole set of CPI items and to the set of CPI items pre-classified as belonging to CPI inflation’s common cycle. We have named them control set and preferred set, respectively. These are alternatives to the set of official classifications (conventional set). The disaggregated models are estimated by using these three sets of groups. We find that: (1) the forecasts obtained by the two sets of homogeneous groups are more precise than those obtained by the conventional set. (2) Those obtained by the control set are the most precise forecasts. Therefore, in order to improve the forecasting performance of disaggregated models, the use of clustered homogeneous groups is highly recommended.