Churn Analysis using ChatGPT

In my experience, there are quite a lot of SaaS founders who are not really data-driven. I am sometimes surprised at how far they have made it with gut instinct.

In most cases, they at least track the most important figures - they just don't use them (enough) in their decision-making processes.

Today I would like to show you how easy it has become to make use of your customer data thanks to AI. You don't have to be a statistician or have a penchant for numbers.

The use case I have chosen for this article is churn analysis, but there are many more:

• Refining your ICP,

• Prioritizing product features,

• Improving onboarding process,

• Identifying upselling opportunities,

• Enhancing efficiency of marketing initiatives,

• …

Please note that I created the sample data set using ChatGPT, so it may not necessarily be representative. Also, I didn't invest a lot of time in prompt engineering, so there is probably plenty of room for improvement.

Read in dataset and describe contents

I started by providing the sample dataset and asked for a description of the contents. As you can see, the dataset contains some information about the customers as well as product usage.

Cleaning dataset (if needed)

In my case it was an already cleaned dataset - 1000 out of 1000 rows remain.

However, if you're doing it with your own data, you shouldn't skip this step. ChatGPT knows what to clean up datasets and carries out the necessary tasks automatically.

Optionally, you can also get high-level summary statistics of variables as shown below.

Plotting data

If necessary, it is possible to go deeper into the data set/individual variables and create useful charts. I have included some simple examples below.

ChatGPT usually uses the most suitable chart style straight away, but you can ask to change chart style/presentation/color scheme if you are not satisfied with the initial result (e.g. if you want to use your company’s color scheme for an investor presentation).

Logistic Regression Analysis

Now to the most interesting part → Conducting a regression analysis to find out which factors influence churn.

I used logistic regression, a simple and efficient ML algorithm to predict a binary outcome (in this case, whether a customer churns or not) based on a given set of independent variables.

Might be worth trying out other methods too, up to you.

The outcome (dependent variable) was coded 0 (didn’t churn) and 1 (churned) in the last column of the data set called “Churn”.

Since we don't want to bother with interpreting statistical numbers we are not familiar with, we ask for an interpretation of the results and advice on possible measures.

Doesn't look that bad, does it? It shows at least some areas where improvements are possible and have the greatest impact.

Still, it's certainly not the best result to get out of it - just wanted to show you that it's worth playing around with.

By the way, what also works great is the analysis of SaaS transaction data (e.g. from Stripe). For example, you can do a proper cohort analysis relatively easily.

I'll write a separate post on this, but if you want to play around yourself, here's a sample dataset from Christoph Janz's blog post on cohort analysis.

If investors or acquirers now ask you for figures and you don't have them to hand, simply upload raw data to ChatGPT and start asking questions.