Subgroup Analysis (SA) is a helpful technique in the context of randomised experiments and in observational studies. With reference to program evaluation, it helps in determining whether and how treatment effects vary across subgroups induced by baseline covariates. However, the choice of the optimal number of subgroups is often ambiguous and causes concern. Here, SA is conducted using the cluster-based approach introduced in D' Attoma and Camillo (2011) and the usage of the Information Complexity Criterion to select the optimal number of groups is proposed. A simulation study and a real case have been illustrated to show such promising approach.