How to use social media to invest in startups

Most startups go bust in the first five years: 78% of them, according to the more methodologically reliable estimates. In fact, even the most experienced venture capitalists lose money on more than half of the new companies they invest in.  This being the case, rather than “invest,” maybe we should use the word “bet.” Even managing partner of Google Ventures Bill Maris compares investing in startups to playing the lottery.

Today, thanks to a new methodology developed at SDA Bocconi, based on analyzing social media and using machine learning (ML) tools, startup investors can improve their aim. Our methodology has already proven it can to get a fix on the most promising new ventures from the moment of launch with greater precision than professional investors.

The reasons behind most failures are that startuppers have a poor understanding of the market and an inaccurate interpretation of market demands. Just consider the typical profile of a startupper:  an innovator who is so fixated on the technical qualities of their brainchild, it never even occurs to them that the market might not be interested.

It is no simple task to figure out whether or not a founder or a group of founders really understands the market. To find out, scientific literature and analysis by investors center on the characteristics of these people, both as far as prior experience and competencies, as well as psychological makeup. But so far these efforts have been in vain, as evidenced by the fact that investors aren’t getting any better at spotting the most promising companies.  Another conceptual obstacle is recognizing the individuals who would make up the potential market for a product that doesn’t exist yet.

The new methodology we mentioned is inspired by the idea that founders who have the same mindsets as their markets will have a better chance of understanding the relevant needs and seizing the resulting opportunities. Social media, in particular X (formerly Twitter), has turned out to be the place where we can discern potential markets and measure the resemblance between the mindsets of founders and these markets by using ML tools (the most advanced AI field to date.)

We analyzed 1,800 startups launched in the US in 2014 and 2015, and for every business sector we identified “benchmark” Twitter accounts, where relevant topics central to the business in question were regularly referenced. For example, if a company’s field of business is fitness, the benchmark accounts might be popular fitness magazines or major gym franchises. We equated the friends of a number of these accounts to the potential market for the startup in question.

To evaluate like mindsets between founders and their markets, on one hand we counted how many of the founders’ friends were also friends of the people we typed as belonging to the potential market (friend similarity). On the other, with the help of an ML tool, we checked how the founders’ language compared to the people in their potential market (linguistic similarity).

The companies we classed with these criteria achieved better results in subsequent years than the venture capitalist picks in terms of profitability, return on investments and lifespan.

First, the companies backed by venture capitalists failed in greater numbers than the ones we identified with the two similarity criteria. More specifically, 23% of the top-30 companies for investors failed in the first five years of business, compared to 10% of the top-30 ranked by linguistic similarity, and none of those that demonstrated friend similarity.

As for return on investment, the top ten for linguistic similarity tallied returns amounting to $2.2 billion in eight years, compared to $420 million for the investors’ top ten. So if professional investors adopted this new methodology, they could spot startups with the best potential in the earliest stages of development, which would make investing far less like gambling.  

Originally published in Fortune Italia