Palantir has a core team of engineers building out a very general data analysis framework. Their core product consists of a GUI for manipulating and sharing annotated graphs (i.e. edges and vertices). The actual edges and vertices of the graph can have complex types defined by some arbitrarily detailed ontology. The GUI also includes tools for producing visualizations and charts from numerical and textual attributes of the edges/vertices. The core product is very powerful when configured properly and it is used to solve important problems in both government and industry.

Palantir also has a much larger corps of consulting engineers (FDEs in Palantir parlance) who actually build the ontologies, import a customer's data in to Palantir, and then provide ongoing support to users. A huge portion of this work is writing scripts to munge the data from arbitrary formats (SQL tables, CSV, Excel, etc) to the format supported by Palantir. A core reason for Palantir's success is that they convince large numbers of top engineering students from MIT, Stanford, Berkeley, Harvard, etc, to do this relatively unglamorous work for below-market pay.

Due to the imbalance in headcount between core engineering and the consulting division, Palantir, in practice, operates more similarly to Accenture or McKinsey than to Google or Facebook. However, for some unknown reason, their valuation seems to be calculated like that of a high margin software company, rather than like that of a low-margin services business. In particular, to increase revenue by N dollars, Palantir needs to hire O(N) employees.