Towards Understanding Retrieval Accuracy and Prompt Quality in RAG Systems