I’ve recently been looking into artificial intelligence and machine learning (AI/ML). Early on I noticed a high proportion of exits via acquisition with relatively few IPOs.
Since 2000 there have been 110 M&A transactions involving AI companies vs. only 6 IPOs. Interestingly, several of these acquisitions occurred very early in the business lifecycle, sometimes even before a product was released. These “acquihires” spurred a thought, which turned into a hypothesis:
Pure play AI and Machine Learning startups make poor VC investments because they tend to exit early, before achieving outsized returns.
To test this I examined CB Insights data on the 221 VC-backed AI/ML companies since January 2010 in search of answers to the following:
- How many AI/ML companies received early stage VC investment? (Seed or Series A)
- Of the companies that secured early stage funding, how many went on to secure growth/late stage VC investment? (Series B+)
- Of the companies that secured early stage funding, how many of these companies achieved an exit?
- Of these exits, how many were acquihires? (For the purposes of this analysis, I’ve defined an acquihire as an exit prior to raising Series B)
I then compared the AI/ML companies to those in other industry verticals: Social, Mobile, eCommerce, Mobile Commerce, Marketplaces, Payments and Entertainment.
I realize that this definition of an acquihire is not 100% accurate. Some startups may reach profitability and not need Series B financing. However, I stand by the generalization that selling a business within five years or less raises question as to whether the startup had the potential to be a stand-alone business. In any event, the assumption has been applied across all verticals in this analysis.
Payments, eCommerce and AI offer the highest likelihood of securing late-stage VC investment after a Seed or Series A round (25.2%, 24.4% and 24.2% respectively). Surprisingly, only 11.4% of Social businesses secured a growth round.
Of the 221 AI/ML startups that received early stage investment, 21 (11.8%) achieved an exit. This 11.8% ratio was the highest across the selected sectors. Of those 21 exits, 15 took place before a growth investment meaning an “early-exit” rate of 8.4%. This too was highest among the compared sectors.
Many factors could be driving this. It’s clear there is an enormous demand for machine learning capabilities to be brought in-house and it’s far easier to buy top talent than build it from scratch. Alternatively, many AI/ML startups struggle to overcome the cold-start wall, which is the obstacle of gathering sufficient training data to teach an algorithm. Companies facing this challenge might be lured early on to large tech companies like Google and Baidu because of their enormous oceans of data.
Of the sectors I examined, Marketplaces were the least likely to exit within five years of an early investment round — only 4.6% did so. This could be due to intense competition in the sector and the recent boom in the sharing economy. It could also be inherent in the business model: building liquidity takes time.
The analysis is interesting but I am reluctant to claim it as conclusive proof of my hypothesis. Across the seven sectors investigated, only three had more than 30 exits, suggesting that the sample size may not be statistically significant. Furthermore, early exits are not necessarily a bad thing for investors. Wit.ai raised only $3M in venture funding before being acquired by Facebook. The amount was undisclosed but I’m assuming the VCs did well.
In general however, early exits at “lower” prices do have implications for the economics of VC investment. In order to make the types of returns that have a meaningful impact on their fund, VCs need to achieve higher ownership stakes at lower valuations. Looking back as far as 2000 only one AI acquisition broke $200M within a five-year window from Seed/Series A to exit: Google’s purchase of Deepmind. The majority of AI/ML exits were $100M or less in value. The conventional wisdom is that VCs must engage in “elephant hunting” — seeking out investments that, on their own, have the potential to return the fund. If this holds, then investing in pure play AI/ML startups might only make sense for micro VCs or angels.
Do these findings surprise you? Have I missed anything? If you have any thoughts on this topic, I’d love to hear them. Please leave comments below or tweet me at @mattwichrowski.
A huge thanks to John for helping me with this article.