Development and demonstration of a ChatGPT model for importing and standardizing financial data

This study develops and demonstrates a practical pipeline for importing and standardizing financial data from agricultural cooperative audit reports, many of which exist only as scanned PDFs. The goal is to enable benchmarking and ratio analysis despite heterogeneous report formats, terminology, and layouts. Using ChatGPT-5 to assist code generation, the system implements a Python workflow that renders PDFs to images, applies optical character recognition with light preprocessing, reconstructs page layout from word positions, and exports a standardized dataset to Excel. A core design feature is “year-band” targeting: the model detects year headers and restricts extraction to the corresponding column, reducing cross-year contamination common in multi-period statements. Key balance-sheet and sales items are captured via label matching. When a printed-total line is absent, a strict section-sum fallback reconstructs the figure by adding line items within that section and target-year band. The pipeline is demonstrated on two Oklahoma cooperatives whose PDFs exhibit differing formats and terminology. Findings indicate high coverage for headline items, reliable metadata extraction (cooperative name and city), correct handling of negative amounts (parentheses), and practical runtime on standard hardware. The approach is transparent, auditable, and scalable to batch processing, enabling the construction of liquidity, profitability, and solvency ratios for research and managerial use. Limitations include sensitivity to scan quality, atypical layouts, and the current focus on balance-sheet variables. Proposed extensions include multi-year harvesting per document, additional income statement and cash-flow items, a native-text fast path for digitalized PDFs, and light computer-vision cleanup (rotation). Overall, the system provides a reproducible bridge from heterogeneous scanned statements to a clean, analysis-ready dataset for cooperative benchmarking and policy-relevant empirical work.