Analyzing Data with GPT4: Are Data Analysts Doomed?

so gpt4 is here and it's the updated version of the model powering chat TPT in this video I'm going to show you how gpt4 is capable of doing actual data analysis and extract useful information from raw data I'm going to show you differences when it comes to data analysis between the old version 3.5 and gbt4 I'm also going to cover who I think is most likely to benefit from this technology and how I would approach this going forward and by the end of this video I hope that I will have convinced you that this is not a fad this is not going away this is going to change the way we work forever alright so we're now going to analyze some data and we're going to start off with the old version GPT 3.5 just to be able to make the comparison and we're going to be analyzing e-commerce data here I have some Shopify data that I have extracted from the Shopify API this is from a development store it's a small data set but enough to get us started so let's have GPT 3.5 analyze this data set and because it's a small data set we can simply paste the raw data into the prompt and we see that chat TPT recognizes this data as order data from a Shopify store and highlights that there are repeat orders in the data set which is nice let's see if chat TPT can find the average order value so chat if G knows how the AV is calculated and that in this case we need to use the subtotal price set and apparently GPT 3.5 has no problem doing the actual calculation behind the aov so let's now have chat Liberty give us a table with all the orders to give us a nice overview since the data that we copy pasted was raw and here we have the first problem so chat DPT 3.5 claims that it doesn't have access to the order data which is clearly wrong next let's try to get the lifetime value of the clb for every customer and we know that we can do that calculation based on the provided data and it seems like gbt 3.5 knows what clv is but then proceeds to give us an answer that is just nonsense we obviously don't need the custom acquisition cars to calculate the clb so this is like a college kid not knowing how to solve a problem in an exam just writing down some basic concepts now let's compare this to what gbt4 is capable of so we're going to start a new chat and select gpt4 and then we're going to paste in the same raw data and have gpt4 do the analysis so analyze the following data and we'll paste in the same data as we did before so chat dpt4 also recognizes this data as order data from a Shopify store which is not surprising and it gives us the same summary of the different columns as GPT 3.5 and I actually think that the summary of the data is more clear with gpt4 than the summary that we got from gbt 3.5 now let's ask gbt4 to put all the orders in a table which GPT 3.5 couldn't do and here we see that we get the table that we want which gives us a clear overview of the raw data that we pasted into the prompt so the service is very slow right now which is probably due to the amount of traffic that open AI is experiencing all right so we got the table that we wanted let's see if gpt4 can give us the clv for every customer and we see that 2b24 is actually identifying that we need the total spending for every customer and then that we need to add this up so we see that gpt4 gives us the clb for for the customers there are actually more customers in the data set I'm not going to bother asking again but this is clearly better than gpg 3.5 so let's try to ask chat DBT how it would predict clv and let's give it a horizon let's say two years and what we see is that gbt4 will actually outline a high level strategy on how this can be accomplished Which models that could be used which data would be valuable I think it's a little bit too General given the fact that it actually has a data set and not specific enough in terms of what models could be used to accomplish this so let's be specific and ask about a model and see what we get and now we get something specific that we can actually use so this is the model implemented in the lifetimes python Library the BG NBD model and the gamma gamma model would be a natural place to start to model the clb for a data set like this and I actually think that dbt4 does a nice job of outlining the different steps that you need to go through to calculate the clb or predict the clb for customers based on order data like this so this is clearly going to be valuable for junior data analyst a data scientist without the experience to be guided in the process of modeling customized term value I've created a video about this I'll put a link to that video below this one that shows how this can be done for a Shopify store but the outline here is almost identical so let's ask gpt4 to actually write the code for this so we don't have to so we can see that the right libraries are being imported it's importing pandas it's importing summary data from transaction data and it's importing the Fitters from lifetimes that will estimate the parameters of the model than the rfm data frame is created which is order data and the format needed to fit the model and then check Deputy goes on to fit the first part of the model which is the beta KU fitter based on the rfm data and this is nice so chat to PC actually understands that to fit the gamma gamma feeder it needs to select only the customers with a repeat purchase once both parts of the model has been fitted chat to petite goes on to predict the number of purchases and the conditional expected average profits which is very close to what we want if you're interested in how to build a model like this I suggest that you check out the video I mentioned before that I will link to below this one overall I think gbt4 seems very promising for junior data scientists and data analysts missing the implementation experience and this will definitely speed up the development of a useful model for more senior data analysts I suspect that the true power of gbt4 will only be revealed once you start using the API and gpt4 in combination with pandas to really speed up the data wrangling process all right that's it for now if you enjoyed this video give it a like And subscribe thanks for watching

Channel: Rabbitmetrics

Views: 11,793

Keywords: gpt4, gpt-4, chatgpt, openai, shopify data, data analyst, data science, clv, data analysis, chat gpt-4, gpt 3.5, gpt-3.5, ecommerce data analyst

Id: 6WE09Ihdn9M

Length: 9min 14sec (554 seconds)

Published: Thu Mar 23 2023