In the realm of Big Data systems, Hadoop has emerged as one of the most popular systems and a very diverse ecosystem has grown around it, meeting all kinds of functional and technical needs. One niche that should have been a place of choice in this ecosystem is data analytics: first because getting value out of large datasets requires efficient Machine Learning (ML) algorithms, second because large clusters with abundant CPUs resources seem like appropriate playfields for ML algorithms which are often very resource-intensive computing tasks. Unfortunately among the myriad of open source projects, there are very few data analytics tools that have been ported to the Hadoop framework. Apache Mahout stands out among those rare initiatives: this project is mainly known for its recommendation application, but it also offers a warehouse of ML algorithms, advertised to run on Map/Reduce. We did investigate the twenty algorithms proposed within Mahout and in this report we focus on the most promising one: the Random Forest implementation. Relying on extensive tests, including specific marketing data from Orange, we provide an in-depth feedback on the use of this tool, both from a practical and theoretical perspective, and we suggest several improvements.