A Guide to Data Analysis with ChatGPT: Public Data Analysis Even Beginners Can Do

Why is it so hard to build an MVP/service on your own? June 16, 2025 You have probably heard from many sources that data analysis matters. The reason we need to do analysis here is that when you build a Product, you create evidence for it, and that is why we carry out data analysis. While building a Product, you may start to doubt your own product and its features, and if you fail to reach the targets you planned (KPIs, etc.), your team's morale and trust can also take a hit. So the main purpose is to reduce early external variables as much as possible. Of course, more than 80% of first Products end in failure, and I want to stress that developing based on users' feedback throughout that process matters even more. We will take a brief look at how to find, process, and analyze data. Before analyzing data, there are a few things you need to set up; I will cover those in a later post. (If you are in a hurry, I recommend asking ChatGPT to suggest the two items below.) For now, I will focus on the point that with ChatGPT, you can do data analysis without writing code yourself. Naturally, you can also check this easily by using GPT. First, I looked through various public datasets and tried to analyze whether they were connected, but most of them differed in "year," "region," and so on, which made it hard to link them in common, and therefore hard to analyze. There are many other datasets available as well, so I encourage you to make use of them. Based on the data downloaded from the Seoul Open Data Plaza above, I made the request shown above to analyze the data. Unfortunately, an error code came up, so I asked ChatGPT about that too, and it very kindly explained why and gave me corrected code. Next, I wanted to know whether the datasets had actually been merged properly, so I asked for code to check that, and fortunately I confirmed that the data had come out well. Looking at the data, the "difference in count" for 2016 was NaN, so I judged it better to exclude the 2016 data and made that change. But another error came up at this point, so I asked GPT to fix it. I wanted to see the relationship between the "difference in count" (the increase in Eco Mileage sign-ups) and the "current-year carbon difference per sign-up" (the increase in an individual's carbon emissions), so I asked the question below. The correlation coefficient came out at "-0.38," so I judged that there is a correlation, though a weak one (correlation coefficients fall between -1 and 1, and the closer they are to -1 or 1, the stronger the correlation). Going a step further, I also wanted to see it as a graph, and got the result below. What I have written here may be difficult for those doing data analysis for the very first time. If you have any questions, feel free to leave a comment anytime :)

---

If you're wondering which tasks you could hand over to AI,
a 2-minute assessment can show you where to start structuring first. Feel free to answer only as much as you're comfortable with.

STAR-T AI Business Operations Assessment →