The increasing complexity of product functionality and manufacturing processes often leads to complex failure modes and reliability problems within the product usage phase. This paper outlines an approach to determine and identify risky parts in product fleets based on cluster analysis with respect to product failure behavior and usage load profile. The theory and application of the approach are shown with the help of a data base of a light electric vehicle (LEV) product fleet in the usage phase. Three cluster algorithms are applied in the case study: hierarchical clustering with a Euclidean distance measurement and Ward Linkage, hierarchical clustering with a City-Block distance measurement and Ward Linkage, and partitioned clustering with k-means algorithm. The impact of the use of these different distance determination methods respectively fusion algorithms is analyzed. In addition, a comparison with state-of-the-art risk analyses using Weibull distribution models with candidate prognosis (sudden death) for the whole population and the subpopulation of the risky parts in the product fleet is conducted. As a result, recommendations for field measures (recall and maintenance actions with regard to prioritization and partial maintenance) are derived and evaluated based on the data analyses.