How Sticky Are ChatGPT Plus Subscribers? Don’t Look to its Churn Rate to Find Out (But it’s Good)

ChatGPT continues to grow at a remarkable rate. After reaching 100 million active users within just two months of its launch, primarily through ChatGPT subscriptions and its enterprise business, it has many more now. Using data through December 2023 from Earnest Analytics, below is a summary of customer acquisition, active subscribers, spend per active subscriber, and total sales for ChatGPT Plus:

ChatGPT is continuing to “cross the chasm”, as customer acquisition initially spiked, then fell, and is now working its way back up again. Active customers and total sales continue to power upwards, as spend per active remains roughly constant at about the $20 monthly cost of the ChatGPT Plus subscription.

This begs the question though – just how long are those GPT Plus subscribers holding on to their subscriptions?

Stronger retention means more revenue in the future, as more of their subscribers today generate revenue for the company tomorrow. It also is an indication of product-market fit, as it means that subscribers are presumably happy enough with their subscriptions to hold on to them for a longer period of time (and to tell their friends about it). Finally, it means higher customer lifetime value (and thus better unit economics), because subscribers bring revenue in over a longer period of time, which means more value per customer.

The overall churn rate is misleading at best…

Given the paramount importance of retention, it is perhaps surprising how little that has been written about subscriber churn for ChatGPT Plus. But what little has been written, besides a short mention here, has focused on questions such as “what is the retention rate?” and, conversely, “what is the churn rate?”

A recent paper of mine with Barak Libai and Verena Schoenmueller shows that these are not the right questions to be asking, because the overall churn rate (and its complement, the overall retention rate) are misleading measures of customer retention.

This is not a theoretical claim – when you look to real transaction data for ChatGPT Plus, this is the case!

Using credit card transaction data from Earnest Analytics over the period from January 2023 through December 2023, inclusive, we can compute the overall churn rate as it is typically defined – the number of customers lost in a given month divided by the total number of active customers from the prior month. This is the resulting figure:

The overall churn trend fell from ~11% in early 2023 to ~4% by the end of 2023.

Are customers really getting that much stickier – by a factor of almost 3x – over such a short time frame?

No.

The customers that ChatGPT Plus was acquiring near the end of 2023 were not that different in terms of their stickiness than customers acquired at the start of 2023. They were roughly the same.

There are a few ways of showing this but one is to look to cohort-specific churn measures. Take, for example, the month-one churn rate, of the proportion of customers who churn the month after they were acquired. This is the resulting chart:

Month-one churn is roughly flat. A regression fit to the data has a trend that is insignificantly different from 0 (i.e., it is insignificantly different from a completely flat line, indicating no trend in month-one churn across acquisition cohorts).

We get the same substantive result when we look to other cohort-specific churn measures. Younger cohorts are not statistically different in terms of their stickiness from older ones.

As we show in the aforementioned paper, ChatGPT Plus isn’t unique – we see this same trend across every young subscription-based business that we study. The overall churn rate almost always tends to go down sharply over time, and yet cohort-specific churn measures are not – in fact, we show in the paper that typically, customers get less sticky across cohorts, not more sticky!

?

… So look to cohort-specific churn measures instead (and for ChatGPT Plus, they look good)

So what should you do, if “the” churn rate (or “the” retention rate) is misleading? Look to cohort-specific measures instead.

Consider, for example, retention curves by cohort – the proportion of customers that are still alive as a function of the number of months that have passed since those customers were first acquired:

One older cohort had stronger than average retention. Some others had weaker than average retention. The solid black line is the average retention curve across the cohorts. As we can see, the younger cohorts – the red ones – tend to have retention curves that are pretty similar to the overall average.

This conveys that retention is very strong for ChatGPT Plus. Around 63% of subscribers are still holding on to those subscriptions 9 months after subscribing. And if anything, the average retention curve is bending upwards: the so-called (and highly desirable) “retention smile” (more here and here).

Another helpful chart is that of cumulative spend per acquired customer as a function of tenure, by cohort:

Blue cohorts are older, red cohorts are younger. Again, there is very little evidence that younger cohorts are all that different from old ones - the red lines are visually indistinguishable from the black line representing the overall average.

But it is also striking just how straight this line is. The average customer generates $156 in revenue over the first 9 months of their life. Considering that the average monthly cost of a subscription is $20, if no one in the cohort had churned, cumulative spend per acquired customer be $200. This isn’t that far off. Again, this speaks to how strong retention is.

In summary, we see using data for ChatGPT Plus what we show in our paper:

1.???? The overall churn rate is misleading at best for conveying underlying trends in retention and customer value

2.???? Cohort-specific measures are better, and tend to be far less sanguine (typically degrading over time). This is why I am a strong advocate for Earnest Analytics, because it provides unique insight into cohort-specific data.

ChatGPT Plus is no exception. That being said, we see that the absolute level of customer retention is very strong for ChatGPT Plus, which is a testament to the value the product creates for its subscribers, signifying strong product-market fit and value being created for users.

[ And as correctly noted by Alexander Izydorczyk, the net dollar retention rate has the same issue because it too is computed at the company-level / is not cohort-specific: https://magis.substack.com/p/churn-paradox ]