Community challenge: Using Make’s AI Content Extractor to process 1,200+ messy purchase orders

• Industry: Professional Services - Business and Management Consulting
• Use case: Order Fulfillment
• Country: Colombia
• Company size: 2-10 employees
• Apps used: Make AI Toolkit, Make AI Content Extractor, Make AI Agents, Google Drive, OpenAI (ChatGPT, Sora, DALL-E, Whisper)

Over in the Make Community, we love setting challenges to see what makers dream up when given a prompt and some creative freedom. For our latest challenge, we invited users to build scenarios using the Make AI Toolkit, the Make AI Content Extractor, or both.

Santiago Ruiz, the Community Challenge contestant and founder of TheMagicHack, submitted an impressive solution to a real-world problem familiar to many businesses: processing large volumes of purchase orders.

His automation creation landed him a finalist spot in the challenge and showcased perfectly how powerful AI can be when paired with Make’s visual automation platform.

The challenge

• One of Santiago’s clients receives around 150 purchase orders per month from 13 different customers.
• Before automation, every purchase order had to be manually reviewed and entered into their internal systems
• Key data points, like item lists, totals, requester info, and delivery details, were buried in documents
• The layout and formatting were inconsistent
• Manual entry slowed down operations and caused frequent mistakes
• The process was time-consuming, error-prone, and hard to scale.
• Santiago wanted to build a solution that could handle the mess and eliminate the manual work.

Santiago Ruiz, founder of TheMagicHack

The solution

Santiago built a scenario using Make’s AI modules that could:

• Watch for new files in a Google Drive folder where purchase Orders are saved
• Extract raw text from each document using the Make AI Content Extractor
• Send the extracted text to OpenAI via the “Create a completion” module to structure the data consistently
• Feed the structured output into a PostgreSQL database used by the client’s internal app to track purchase order status

Thanks to this setup, purchase orders are automatically processed the moment they’re received – regardless of layout or source.

Taking it further: Adding an AI Agent for validation

After seeing the initial success, Santiago enhanced the solution by adding an AI Agent that validates and cross-checks data before it reaches the database. The Agent verifies EAN (European Article Number) codes, quantities, and other critical information, applying business rules and automatically catching inconsistencies.

When corrections are needed, the Agent calls specialized tools – one to complete missing EAN data and another to submit validated information to the database. This self-correcting layer ensures only 100% verified go through, making the entire workflow fully autonomous.

The results

Since going live, the scenario has processed 1,800 purchase orders and counting.

Here’s what it’s delivered so far:

• 99.7% accuracy in data extraction
• Zero manual input needed to process incoming purchase orders
• Full purchase order history backfilled from January 2025
• Structured data now flows directly into the client’s operations app
• Sales, inventory, and logistics teams all have access to up-to-date purchase order data
• Self-correcting validation that catches and fixes data issues before they enter the system

Final thoughts

The Make Community Challenge was all about turning smart ideas into real, working solutions, and Santiago nailed it. His scenario didn’t just meet the brief; it proved how accessible and effective AI-powered automation can be.

Best of all, the results have inspired the client to automate even more.