I Summarized My Experience Automating Data Analysis with the ChatGPT Code Interpreter
Every Monday morning at work, I used to organize weekly sales data in Excel and summarize it into a PPT. I automated this with the ChatGPT code interpreter. Let me walk you through the process.
1. First, I write a query to export raw data as CSV. You should do preprocessing and deduplication directly in SQL to keep the file size small.
2. I enter this into ChatGPT: "Analyze this CSV and give me the weekly sales trend, category breakdown, and week-over-week change rate." Then, if I need additional metrics, I add them one by one through the conversation.
3. Once I get the results, I ask it to save them as an Excel file, and then immediately ask for a PPT slide script. The key here is to request insights, not just numbers. If you copy the numbers as-is, the slides end up as plain tables.
4. Finally, when moving everything into Figma or PowerPoint, I just set the tone and keep the graph colors consistent.
This reduced my Monday morning task from 40 minutes to 5 minutes. Of course, it took time at first to understand the data, but once you build the pipeline, it's really convenient. If you're interested, try it on a similar task.
7 answers
Agreed, cutting down Excel work is legit a godsend.
I recently automated something similar, and the key is input data consistency. If the source data has nulls or mixed units, the results can get weird, so it's better to validate and run the preprocessing script first. And when setting the final PPT tone, I find it easier to maintain by defining a template and only making variations within it.
Well, our company has blocked the code interpreter itself due to security and privacy issues... Are there any places that actually approve putting company data into external AI? Must be nice for those that can do it.
Oh, I should try this out too.
Honestly, isn't 40 minutes → 5 minutes a bit of an exaggeration? If you factor in the initial setup time, that wouldn't be the case. And once you automate it, you also need to consider the time spent adapting when the data format changes later. Still, it does help in the long run.
How accurate is it when you ask GPT for insights? Doesn't it sometimes fabricate plausible-sounding nonsense?
Thanks for the great tip. When I make graphs in the code interpreter for my weekly reports, I always set the font explicitly because of Korean font issues. And for PPT, I've found it more reliable to prepare a pptx template file in advance and fill it in rather than using python-pptx. Just a heads-up.