Thread by @breadcrumbsre

Thread on Palantir $PLTR

Investors met Palantir's S-1 with scepticism. It's really hard to be excited when a 17 y/o company has such financials:

Yet the company is very unusual in many senses which led me to do some work on it

On the surface, S&M expense looks extremely high. Yet assuming a very long customer life (top-20 customers have been with Palantir for 6.6 years on average) and taking into account recent improvements in customer onboarding - there is a chance unit economic might work.

The other condition needed to make unit economics to work - you need a very low customer churn. And this is where I stuck.

Palantir does not disclose customer churn in S-1, yet we can get some color of that key metric.

Based on my math they've lost around 8 customers in H1 2020 (comparing end of period 2020 customers with 2019 average)

Yes, H1 2020 was not a typical period because of COVID. But what does this say about stickiness? And many software companies increased their customer count during this period.

Another data point comes from this list of European customers of Palantir algorithmwatch.org/en/story/palan…

According to this source, by Nov 2019 around 49 entities accross Europe had used Palantir, out of them 27 had churned by that time and status of another 8 was unknown. Losing over half of your customers even over an extended period of time looks pretty bad.

On top of that there were some reports on lost partnerships with blue chip companies like Coca-Cola, Home Depot, JP Morgan and American Express as well as with some police departments (LA, NY etc).

Meanwhile, Palantir has failed to diversify their revenue beyond key clients: share of revenue from top-20 customers has been steady at around 70% over the past 6 years

While their economics looks interesting on the surface, it can work only with extremely low customer churn. Yet, I can't get comfortable with data points on churn.