Space Syntax research has shown how human behaviours in the workplace are shaped by spatial configuration; in turn, evidence-based design practices have highlighted ways in which this data can be used to inform tailor-made solutions in office design. Yet, existing research focuses on either single case studies or comparisons of a few cases on a small scale. Also, each study uses its own methods and metrics which makes it difficult to establish wider patterns beyond single datasets. This paper presents a larger than usual data set on workplaces, which has been collected by Spacelab, a design and consultancy practice based in London. This dataset includes spatial and space usage information such as syntactic analysis and desk occupancy on client companies. It resides in a spatial relational database, allowing for systematic combination of the collected data, useful for doing either deeper analysis, or generating benchmarks and baselines. These insights are not only highly relevant to clients but also give rise to opportunities to generate new insights on office design and human behaviours in the workplace from a research perspective Two main research questions relating to the size of samples are discussed: Firstly, whether large samples are necessary to fully understand phenomena, and secondly, whether behavioural patterns vary across cases. Observation data and syntactic analysis are combined to understand in which areas of an office different activities take place. Observation data is also brought together with the functional allocations of space in order to ask whether activities follow the programme introduced by functions such as meeting rooms, kitchens, workspaces, etc. It is shown that observation data only becomes robust and reliable with longer periods of observations than previously recommended. Three to four full days seems to produce reasonably stable results for desk occupancy, while five full days seemed required for percentages of people walking and interacting. Some surprising findings were revealed regarding the distribution of activities in space, for instance dispelling the myth that interactions happen in corridors and highlighting that interactions tend to occur in rather segregated spaces. While it is argued that predictive power of the analysis varies, first steps towards establishing generic patterns have clearly been taken.