Utilizing Artificial Intelligence to Drive Operational Efficiencies in Retail Through the Transformation of Unstructured Invoices into Structured Data

The retail sector generates substantial unstructured data, especially in the form of invoices with varying formats. This paper examines methods for converting such data into structured formats by comparing traditional rule-based approaches with large language models (LLMs). The system combines Optical Character Recognition (OCR) for text extraction and evaluates multiple postprocessing techniques like regular expression (Regex) parsing, Named Entity Recognition(NER), Part-of-Speech (POS) tagging and LLM data processing. The comparison focuses on each method's performance in handling both predefined and new invoice formats, considering factors such as accuracy, generalization, computational efficiency, and scalability. The results provide understanding into the strengths and limitations of each approach and offer guidance on selecting the most appropriate method for different invoice processing needs for each practical use case. This study provides a framework for decision-making in the automation of invoice data extraction, helping businesses choose methods based on their specific requirements and available resources.