I’ve been looking for a structured way to use AI for data analysis, and I just found a system that completely changes the approach. It’s about moving beyond simple prompts and creating a reliable, repeatable workflow. I stumbled upon this awesome video from an ex-Meta data scientist, and the creator breaks down her entire process for what she calls “vibe analyzing.”
She took eleven courses on the topic to distill these methods, and her insights are pure gold. The core idea is that to get trustworthy results from AI, you need a plan. She introduces two key frameworks: one for deciding when to use AI and another for how to approach the analysis itself.
The Two-Framework System
First, there’s the ACHIEVE acronym, which helps you identify the perfect use cases for AI in data analysis. It covers everything from aiding team coordination to scaling your best ideas. Second, and this is the part that really got my attention, is the DIG framework. It’s a three-step process for working with your data to ensure accuracy and unlock deeper insights.
Here’s a deeper look at the key concepts the creator shared.
📌 The DIG Framework: Your AI Analysis Blueprint
This framework, which stands for Describe, Introspect, and Goal-set, is the foundation for getting reliable results. The creator emphasizes that you can’t just dump a file and expect magic. You have to guide the AI, and this is how.
- Describe: This is the most critical first step. Before any analysis, you have to ensure the AI understands your data correctly. The expert treats the AI like a very competent but junior analyst. You start by asking it simple things: “List the columns in this spreadsheet,” “Show me a sample of the data in each column,” and “What do you think each column represents?” This initial conversation helps you catch parsing errors or misunderstandings right away. For example, the AI might see “nan” in a salary column and you can then clarify if that means the data is missing or if it’s a parsing mistake. Skipping this step is how you get hallucinations and flawed analyses down the line.
- Introspect: Once the AI has a handle on the data, you move into a more collaborative phase. Here, you ask the AI to think about the data’s potential. A great prompt the creator suggests is: “Tell me some interesting questions that could be answered with this dataset and why they would be interesting.” This does two things. First, it can reveal new angles of analysis you hadn’t considered. Second, it acts as another check on the AI’s understanding. In the video, the AI asked if there were different currencies in a dataset. The creator knew all the data was in USD, so she corrected the AI. This back-and-forth refines the AI’s context and prevents it from chasing phantom patterns.
- Goal-set: Finally, you have to be explicit about your objective. A vague command like “analyze this data” will get you a vague and useless response. Instead, you need to provide clear context. The post’s author gives a great example: “My goal is to answer these specific questions and turn them into an interesting report to post on LinkedIn.” This tells the AI the desired output format, tone, and audience, which dramatically improves the quality of the final result. If the goal was a formal report for your boss, the outcome would be completely different.
💡 Beyond Spreadsheets: Unlocking Advanced AI Capabilities
This is where AI data analysis goes from being a convenience to a superpower. The creator highlights several things you can do with modern AI tools that are incredibly difficult with traditional methods like Excel or SQL alone.
Intelligent Filtering: Generative AI can use its broad world knowledge to interpret and filter data without explicit labels. The example given was a job dataset with only city names. A user could ask to see jobs “on the East Coast,” and the AI would know which cities like New York or Boston qualify. This natural language querying adds a powerful layer of semantic search to your data.
Multimedia & File Automation: The analysis isn’t limited to text and numbers. The video shows how you can upload a video and ask an AI to extract frames at specific intervals, convert them to grayscale, increase contrast, and even compile them into a GIF. Another mind-blowing trick is using zip files. You can zip up an entire folder of disorganized documents, upload it, and ask the AI to read each file, summarize it, propose a new folder structure, and rename the files logically. Then, it can zip it all back up for you to download.
Conversation-to-Code: This one is absolutely awesome. After you’ve gone through a series of analysis steps in a chat, you can ask the AI to turn that entire process into a reusable program. The creator showed a prompt like: “Turn this process into a Python program that I can download and run on my computer.” The AI then generates a script that automates the whole workflow. This creates a “traceability document,” allowing anyone to replicate and validate your analysis. This solves a massive problem with reproducibility in data work.
✅ From Analysis to Application: Making Your Data Work for You
The final point the creator makes is that the analysis doesn’t have to be the end of the road. The insights you generate can be the starting point for creating tangible assets and even full-blown applications.
The output from your analysis can be directly transformed into a variety of formats. You can ask the AI to generate a report, a PowerPoint presentation, a series of social media posts, or even personalized emails. The video revisits an example of a workshop where the AI analyzes an attendee list and then generates custom cheat sheets for each person based on their role and interests.
Even more powerfully, you can build entire applications on top of these analytical workflows, often with no-code or low-code methods. The creator calls this “vibe coding.” An analysis of traffic data could become a real-time alert app. An analysis of investment data could be turned into an AI agent that provides research on demand. This shows a clear path from a simple data query to a scalable, interactive product.
This video really opened my eyes to the potential of structured, thoughtful AI data analysis. It’s not just about getting quick answers; it’s about building a robust process to get reliable and creative results.
If you work with data at all, you have to see how this talented creator lays it all out. Check out her full video for the specific prompts and detailed walkthroughs!