Imagine teaching a computer to read and understand complex business documents like invoices, contracts, or receipts. That's the challenge of Business Document Information Extraction (BDIE), and new research shows how AI is getting remarkably good at it. Traditionally, extracting key data from these documents has been a tedious manual process or required complex, rule-based systems. But what if AI could learn to use the software tools we already have, just like a human employee? That's the core idea behind Retrieval Augmented Structured Generation (RASG), a novel approach that treats information extraction as a "tool use" problem. By combining smart retrieval methods, supervised fine-tuning, and structured generation, researchers have built AI models that can extract key information and line items from documents with impressive accuracy. These models, even relatively small open-source ones, are outperforming larger, more complex AI systems. One key innovation is the way these models are "taught" to use tools. By structuring the input prompt to resemble the original document layout, the AI can better understand the context and extract the right information. This approach also allows the AI to adapt to new document types and tools without extensive retraining. The research also introduces a new metric, GLIRM, for evaluating line item recognition, which is more aligned with real-world business needs. Plus, a clever algorithm helps pinpoint the location of extracted data on the document, even without visual analysis. While there are still challenges, like handling variations in document layouts and improving the accuracy of bounding box calculations, this research opens exciting new possibilities for automating document processing. From streamlining financial operations to accelerating contract analysis, AI-powered BDIE has the potential to transform how businesses handle information.