A data scientist for the price of a Netflix subscription

This article is just an extract from my main weekly newsletter Tech Pizza. You should join it here.

OpenAI?released?the code interpreter feature for all their paying ChatGPT users.?

Basically, ChatGPT has now access to its own computer and?can write and EXECUTE code?on it if it’s necessary to answer your questions.?

The main application I’ve seen around is for data analysis, and so obviously I had to try it.?

I took some analytics data from Tech Pizza (open rates, click rates, sources, etc.), anonymized it by removing all email addresses, names and other identifiable information, and uploaded to ChatGPT. I then asked it to act as a mix between a marketer and a data analyst and help me grow the newsletter.?

I prompted it to start by identifying key questions I should ask myself, and it returned these three:?

• Subscriber Acquisition: What are the most effective acquisition sources and channels for gaining new subscribers??
• Subscriber Engagement: How are different subscribers performing in terms of engagement (open rate, click rate)??
• Churn Analysis: What characteristics do inactive subscribers share??

Within one minute I had some pretty amazing results to look at. Here’s the simplest one: a bar chart of subscribers by acquisition source.?

You can see how the paid ads campaigns we’ve run are working ???

Here’s a more sophisticated one, looking at the different engagement rates for different channels:?

Our open rate is around 50%, but I had no idea that the highest engagement was coming from people that discovered the newsletter from Instagram (IG and stories)!?

I pushed it even further and asked to make a linear forecast of our subscriber growth in the next year and say which assumptions it had to make in order for it to do it. It assumed a 20% improvement in growth rate and a 10% reduction in churn, and made this plot:?

also pushed it a bit, running another example on a data science challenge that I shared on LinkedIn, you can read it?here.?

What shocked me, in that case, is that the data wasn’t very clean and had a few problems that ChatGPT had to deal with. I was expecting it to fail…but it didn’t. It correctly made some assumptions on outliers, cleaned the data up, and produced a great analysis.?

I had a chat with a data scientist I’m coaching and we reflected on the fact that some of the junior people working for his team would need a couple of days to do this analysis. ChatGPT made them in less than 5 minutes, for $20/month.?

So how do I feel now that AI is coming for my old job as a data scientist? ???

This is a prime case in which I have conflicted emotions. From one point of view, this is yet another example of how many jobs this technology could make almost completely redundant. People are fighting over the internet about whether Data Scientists are out of a job or not, and I don’t know what’s the correct answer to that question. One thing is for sure: if something that took 2 days to a person now takes 5 minutes to a machine,?that person’s job needs to change.?How? Hard to say, but we can’t ignore it and pretend everything will be the same, because it won’t.?

The other consideration is that these skills were once available to a tiny part of the population, which was fairly expensive to hire.?How many datasets exist in the world? A lot. How many data experts are available to analyze them? Definitely not enough.?So - as usual - we have one scary question (what’s the future of work?) and one very promising outlook for the future in which?everyone has access to once very limited capabilities, for the price of a Netflix subscription.

I think the best way to end this newsletter is with one of my favorite quotes about tech:?

The future is already here, it’s just not very evenly distributed -?William Gibson

So…stop ignoring the future. Help build it.

If you liked this article, you'll love my main weekly newsletter Tech Pizza. You should join it here.