The progenitors of high-mass stars and clusters are still challenging to recognise. Only unbiased surveys, sensitive to compact regions of high dust column density, can unambiguously reveal such a small population of particularly massive and cold clumps. Here we study a flux limited sample of compact sources from the ATLASGAL survey to identify a sample of candidate progenitors of massive clusters in the inner Galaxy. Sensitive mid-infrared data at 21-24 $\mu$m from the WISE and MIPSGAL surveys were explored to search for embedded objects, and complementary spectroscopic data were used to investigate their stability and star formation activity. Based on such ancillary data we identify an unbiased sample of infrared-quiet massive clumps in the Galaxy that potentially represent the earliest stages of massive cluster formation. An important fraction of this sample consists of sources that have not been studied in detail before. Comparing their properties to clumps hosting more evolved embedded objects, we find that they exhibit similar physical properties in terms of mass and size, suggesting that infrared-quiet massive clumps are not only capable of forming high-mass stars, but likely also follow a single evolutionary track leading to the formation of massive clusters. The majority of the sources are not in virial-equilibrium, suggesting collapse on the clump scale. This is in line with the low number of infrared-quiet massive clumps and earlier findings that star formation, in particular for high-mass objects is a fast, dynamic process. We propose a scenario in which massive clumps start to fragment and collapse before their final mass is accumulated indicating that strong self-gravity and global collapse is needed to build up rich clusters and the most massive stars.