Background. Recently, with the high-speed development of Generative AI (GenAI), it is becoming more feasible to automate tasks that go beyond coding, like those usually performed by Product Owners (POs), such as developing feasibility studies, managing requirements, and handling communication with stakeholders. Despite the excitement, there is not a whole lot of concrete understanding about how effective this type of automation actually is today in practice, or what sort of technical challenges and limitations come up when trying to implement it.

Objectives. This research explores whether it is realistically possible to automate higher-level, non-coding tasks that POs at Ericsson are responsible for, using GenAI when it fits. The tasks to be automated include drafting documents, generating emails, suggesting parts of documents such as business backgrounds, and reviewing language. The goal is also to get a better picture of the technical challenges and limitations that appear during the development of the automation tool for POs, and to evaluate how accurate these automations are, along with how much time they can potentially save.

Methods. Several methods were used in this study. An automation tool was built and tested in a practical setting. At the same time, semi-structured interviews were conducted with professionals with hands-on experience with automation, helping to validate and understand the challenges found by us. A survey was also shared with POs to quantitatively assess how automated results compared to traditional manual efforts. Thematic analysis was used to interpret the interview data.

Results. The automation tool reduces the time required to complete certain tasks, often by more than 75\%, especially for tasks such as language refinement or generating initial drafts. On the other hand, tasks that needed company-specific knowledge showed mixed results, mostly due to the fact that the GenAI models do not always have all the necessary internal organization-specific context. Some key issues we ran into included difficulties processing varied document formats, losing important information in images, model hallucinations, missing proprietary knowledge, and slow response times caused by large prompt sizes or model limitations.

Conclusions. Overall, GenAI shows great potential when it comes to helping POs handle their non-coding tasks. However, there are still clear technical challenges that cannot be ignored, especially when it comes to accuracy and making sure the AI "understands" company-specific content. More work should be done testing strategies, such as Retrieval Augmented Generation (RAG), to help make the models more reliable and context-aware.