Patent data has been an obvious choice for analysis leading to strategic technology intelligence, yet, the recent proliferation of machine learning text analysis methods is changing the status of traditional patent data analysis methods and approaches. This article discusses the benefits and constraints of machine learning approaches in industry level patent analysis, and to this end offers a demonstration of unsupervised learning based analysis of the leading telecommunication firms between 2001 and 2014 based on about 160,000 USPTO full-text patents. Data were classified using full-text descriptions with Latent Dirichlet Allocation, and latent patterns emerging through the unsupervised learning process were modelled by company and year to create an overall view of patenting within the industry, and to forecast future trends. Our results demonstrate company-specific differences in their knowledge profiles, as well as show the evolution of the knowledge profiles of industry leaders from hardware to software focussed technology strategies. The results cast also light on the dynamics of emerging and declining knowledge areas in the telecommunication industry. Our results prompt a consideration of the current status of established approaches to patent landscaping, such as key-word or technology classifications and other approaches relying on semantic labelling, in the context of novel machine learning approaches. Finally, we discuss implications for policy makers, and, in particular, for strategic management in firms.