Predicting accurate travel times from the OSM road network requires accurate costs for roads and intersections (ways and nodes). Speeds on ways and delays for left-hand turn (for instance) differ by time and location. We present a general statistical framework for evaluating costs from simple OLS regressions. Our method is scalable in terms of data volumes and contextual precision, from entire way types down to individual turns. It is also fast and naturally lends itself to “standard” tools for OLS model evaluation. The estimated parameters – times at intersections and on ways – are critical for individuals running routing engines such as OSRM.