3 min read

😡 What is the real survival rate of startups?

😡 What is the real survival rate of startups?
Photo by Sigmund / Unsplash

I often feel I'm the only one surprised that no one really cares about monitoring the real survival rate of startups.

Most public incubators in Europe will tell you they have around 20% or less failure after five years. But in 2012 a study financed by the European Commission was giving more or less 3 years of half-life for any startup; which is difficult to reconcile with the first estimate. Asking VCs is not helpful either. Given that they tend to be very skewed to the best projects in the market (which already show traction), and that they exit startups that are not promising enough rather fast, their failure to survival rate is a poor sample of the market.

In a rational world, I suppose that the answer should be coming from public actors like the BPI in France, or the HTGF in Germany. But public investors are usually way more interested in promoting what that kick off than what they actually deliver.

I was interested last week in a post from Lloyd Watts on artificial intelligence startups. He was referring to a 2015 TED Talk listing about 300 deep learning startups at the time, and after some quick checks, Lloyd found sobering results.

Lloyd Watts on LinkedIn: In 2015, Steve Omohundro gave a TED Talk about Artificial Intelligence | 27 comments
In 2015, Steve Omohundro gave a TED Talk about Artificial Intelligence, saying that over 300 Deep Learning startup companies had been funded with a total... 27 comments on LinkedIn

He's listing:

  • 1 IPO at a respectable e$1.5b market cap.
  • 3 acquisitions from $400M to $2b between 2016 and 2019.
  • 7 more survivors recently funded at valuations ranging from $15 to $700M.
(...) it seems the most of the companies have either died, or are Zombies (walking dead), and a small number of them have either gone public or been acquired.

If we extrapolate a little bit, we could trust that Loyd didn't miss any big-name survivors but probably missed quite a few minor ones (he doesn't pretend otherwise). Following this, we could speculate that the real numbers might be 1 well-identified major hit (Sound Hound), 5-10 big successes (βͺ† $500M valuation as a proxy for 'success'), 10-30 solid companies (βͺ† $50M), many zombies still running from grant to grant but not doing any business, and many more dead in the water.

For reference, this is on par with what a typical VC would expect from a specific startup ecosystem:

How 500 Startups described a 'healthy' startup ecosystem only a few years ago. 

Now, yes, I know that these back-of-the-envelope calculations come with tons of caveats, but the thing is: AI is an interesting sample. It's a well-identified mature technology. It deals with a huge number of market applications (both B2B and B2C). And as such, it's oftentimes at the top of the most founded type of technology.

Now, what to make of this?

We're in 2022, and we still don't have metrics to monitor the survival rate of startups. Part of the discussion should be about public money. All over Europe dozens of billions of public money are invested in admittedly risky projects, and that's a very good thing. But shouldn't we be learning about what we are doing a bit more than we are? Ten major successes out of three hundred projects launched is an amazing return on investment in this kind of game.

But beyond finance, it's also about our future capabilities and the development of cutting-edge sectors that could actively boost the fight against climate change, or allow us to better prepare for new pandemics.

If we're ready to invest so much, shouldn't we care a bit more about the outputs?