Bank Statement Extraction via PDF to Markdown

This workflow helps anyone needing to extract data from bank statements, especially scanned PDFs. It is great for accounting departments, financial analysts, or anyone who handles lots of financial documents. This workflow converts bank statement PDFs (even scanned ones) into markdown text. Then, it uses AI to pick out specific information, like all deposit transactions, from the converted text. This makes it easy to get key data from hard-to-read documents. Start the workflow by clicking "Test workflow". The workflow downloads a sample bank statement from Google Drive. You can change this to use your own bank statements. The workflow then uses a third-party service to turn the PDF into images. It uses Google Gemini to convert these images into markdown and then extract the data. You need an n8n instance and access to Google Gemini (or another VLM). You also need a way to convert PDFs to images, such as the Stirling PDF webservice or a self-hosted instance. It uses Google Drive for the example, but you can swap that out. You can change the source of the bank statements (e.g., webhook, other cloud storage). You can also switch the AI model to GPT4o or Claude, though Gemini is recommended for this task. You can adjust what data the AI extracts by changing the prompt in the "Extract All Deposit Table Rows" node. Key Technologies: n8n,Google Gemini (Vision Language Model),Stirling PDF,Google Drive Automation Value: Automatically extracts data from bank statements, saving manual data entry time. Processes scanned PDFs, which traditional OCR struggles with, expanding automation possibilities. Converts complex document layouts into structured markdown, making data extraction reliable. Reduces errors by automating data extraction from financial documents. Allows quick analysis of financial transactions, such as identifying all deposits. Best Practices: For sensitive data, self-host the PDF conversion service (e.g., Stirling PDF) instead of using public services. Optimize image sizes before sending to Vision Language Models (VLMs) to manage token usage and speed up processing.