Microindentation and nanoindentation tests are very useful to assess the micromechanical properties of materials. However, statistical indentation tests require large matrices of indents for heterogeneous materials, which can be time-consuming. Given that specific properties of only one phase of interest can be looked for, a novel procedure is proposed to locate indents automatically. The procedure uses an ε–greedy strategy to determine the next indents to perform. The strategy is informed by the nature of predicted indents using unsupervised clustering of the indentation curves and Gaussian Process Classification. The influence of several parameters has been assessed, and the best combination performance has been quantified in the case of the microindentation of a cementitious mortar sample (500 mN) and the nanoindentation of a cement paste (1 mN). The proportion of indents in the phase of interest increased by around 20% in both cases, paving the way for faster indentation experiments.