Complex manufacturing systems are challenging to study because of the high level of information required and the inaccessibility of most of it. Their tractability is however essential for the efficiency of state-of-the-art industries. This is particularly the case in the semiconductor industry that faces high mix and low volume conditions, and for which traditional methods fail to capture the high complexity and require continuous actions and corrections to adjust to heterogeneous toolsets and product-mix. We present the Concurrent WIP (CWIP), a new way of studying such systems at the level of a process-cluster by identifying each job's queue from its own perspective. CWIP is designed to be practical, with a low level of resource investments, yet informative. We explain how CWIP can be computed based on historical data and then used to derive capacity estimates and clearing functions without any assumptions on the system or on the form of the functions. In the process, we derive not only an average workload-dependent capacity, but also a confidence interval on this capacity. The relevance and efficiency of the proposed estimates are experimentally tested on a simulated system mimicking a small but complex process-cluster of the semiconductor industry. The estimates are used to predict WIP absorption times and we show how they characterize well not only the average behavior but also the full range of possible behaviors of the system. Finally, we discuss further applications of CWIP, that could be used to compute refined clearing functions or to monitor complex systems.