AI Agent n8n: Analyze Startup Pitch Decks

The workflow begins by connecting to your Google Drive and retrieving all pitch deck files from a specific folder. This node acts as the trigger point, scanning your designated folder and creating a list of PDF files that need to be analyzed. Think of it as your automated inbox scanner—it finds every deck waiting for review.

This search operation is configured to return all files at once, meaning you won't miss any decks regardless of how many are in the folder. The workflow will then process each one systematically.

💡 Tip: Create a dedicated folder in Google Drive for incoming pitch decks. This keeps your workflow organized and prevents accidental analysis of unrelated PDFs.

This Loop node controls how the workflow processes multiple pitch decks. Rather than attempting to analyze all files simultaneously (which could overwhelm API rate limits and cause failures), it feeds them through one at a time. This ensures reliable, consistent processing regardless of whether you have 5 or 500 decks in your folder.

The batch size of 1 means each pitch deck gets full attention before moving to the next. This is crucial for AI analysis workflows where you want complete, uninterrupted processing of each document.

💡 Tip: If you're processing a large backlog and want faster throughput, you can increase the batch size to 2-3. However, keep in mind your OpenAI API rate limits—running too many parallel requests can trigger throttling.

Now we fetch the actual PDF file from Google Drive. This node takes the file ID from the previous search results and downloads the complete binary file. The expression syntax dynamically references whichever file is currently being processed in the loop, ensuring each deck gets downloaded in sequence.

This step transforms a simple file reference into actual content we can work with. The downloaded PDF is stored in n8n's binary data format, ready for text extraction in the next step.

💡 Tip: Make sure your pitch decks are actual PDFs with searchable text, not scanned images. The text extraction works best with native PDF documents. If you're dealing with scanned decks, you'll need to add an OCR step.

This node converts the downloaded PDF binary file into raw text that the AI can analyze. It's the critical bridge between a file sitting in Google Drive and actionable text data that GPT-4 can process. Without this step, the AI would have no content to analyze.

The extraction pulls all text content from every page of the pitch deck, preserving the information while stripping away formatting, images, and design elements. What you get is pure textual content—executive summaries, market data, team bios, financial projections, and everything else founders include in their decks.

💡 Tip: The quality of your AI analysis depends heavily on text extraction quality. If you notice missing information in the final output, check the extracted text to see if the source PDF has complex layouts, embedded images with text, or unusual formatting that might cause extraction issues.

This is where the magic happens. The AI Agent node sends the extracted pitch deck text to GPT-4 with a carefully crafted system prompt that transforms it into an expert VC analyst. The AI doesn't just summarize—it extracts structured data, evaluates the opportunity across multiple dimensions, identifies red flags, and generates actionable recommendations.

The system prompt is the secret sauce here. It instructs the AI to think like a seasoned investor, extracting everything from basic startup info to nuanced qualitative assessments. The output parser ensures the response comes back in a structured JSON format that can be directly mapped to your Google Sheets columns.

💡 Tip: The system prompt included in this workflow has been refined through hundreds of deck analyses. However, you can customize it to match your specific investment thesis—add criteria for specific industries, adjust scoring weights, or include questions relevant to your due diligence process.