The increased use of electric vehicles (EVs) to deliver goods in the last mile of the delivery process means that there is a need for more realistic energy consumption models. Benchmark instances of the electric vehicle routing problem (EVRP) currently use an energy consumption model of one unit of energy for every one unit of distance travelled. The underestimation of energy consumption can lead to problems in the routes, in the life cycle and cost benefit analysis of EVs, and their demand and impact on the grid. We use a structured approach to enrich current benchmark instances, following the CRISP-DM data mining framework. This paper provides insights into explanatory variables that impact energy consumption, and provides a replicable methodology to create realistic benchmark instances using real world data.