CNH is a global capital goods company of more than 37000 employees, specializing in equipment and services for Agriculture and Construction. CNHi vehicles cover a vast variety of applications, operating conditions, and environments, making design for reliability one of the main objectives of the company. Warranty claims analysis is a key activity for product support and product development technicians. A widely used source of information is a set of documents, referred to as “Claim Comments” that are filled in by mechanics during repairs. Repair records include valuable information as failure modes descriptions, diagnosis and troubleshooting, potential causes, and other useful details. The information is then fed back for customer/field support and continuous product improvement. Claims details review is mostly executed manually, and engineers need to read thousands of lines each year. This task is constrained by human limitations - e.g., ability of consistently identifying failure modes and the limited number of documents that can be processed at once. This paper describes a solution to make that kind of massive documents analysis more scalable and efficient. Natural Language Processing (NLP) is not a new discipline in data science; however, the problem was tackled in an innovative way using Artificial Intelligence (AI) and a semantic approach instead of the classic frequentist one. This means that the analysis is not limited to single words occurrences but relies on the whole text context understanding. The core technology is based on a deep neural network called BERT [2] (Bidirectional Encoder Representations from Transformers) which can read and understand the sentences, their context and enable the identification of the main topics in a set of documents. This solution can build topics-failure modes relationships, fostering an automated cross-system analysis previously unattainable. The developed application connects to the Enterprise Data System (EDS) through workflows embedded in a cloud-computing environment. The cloud environment allows managing large amounts of raw text, develop code, build data pipelines, schedule jobs, and manage distributed computer resources. The outcome is then delivered through dashboards that can easily convey the model's complexity in an interactive platform. Through the dashboards, users can analyze topics/failure modes across thousands of claims in a more efficient and effective way. The modular structure grants flexibility to the process, which is now able to adapt to different contexts and data sources. Main strengths of the developed solution are the automation and increased productivity for product support technicians, faster failure modes identification and prioritization and easier data exploration, presentation and sharing.