Musings on being fast, early, and machine learning
Fast like Tiger
The subject of VC chatter continues to be Tiger Global, the crossover fund sweeping the industry. I initially thought that TTT (Two-day-Tiger-Term-sheet) was based on exaggerated anecdotes but data shows that Tiger has actually been investing in a new company every 1.7 days. One day to meet the company and send a term sheet (prepared beforehand). Another half-day to agree on revised terms. A friend from another crossover fund lamented how Tiger is kicking their assess - and making them work banking hours. Diligence work that used to take several weeks has to be completed in one.
While Tiger issues a term sheet after the first meeting, this does not mean their due diligence is sloppy. With $80B in AUM, they hire SWAT teams of consultants to help do the diligence and write 80-page investment memos. So it is no surprise Tiger is one of Bain's largest customers. With an army of consultants doing all the work versus a few individuals in most traditional venture funds, Tiger can be both fast and diligent.
While we have been talking about Tiger, the meta point is speed. (The other Tiger-driven lament is valuation, which I wrote about here). My peers constantly talk about it and I have lost opportunities to partner with amazing founders because I was not fast enough. To react faster, one tactic that became common is to outsource parts of the diligence, expert calls in particular. Expert networks like GLG, AlphaSight, and Tegus can set you up with an expert in one to two days. The status quo was pinging a friend to ask if their friend is open to talking...which can take a week...then another week for scheduling...then reschedulings. You get the picture. Expert networks are in so much demand that one of their managing partners told me that they cannot take in any new clients because they cannot hire fast enough to service clients. They turn down requests from small funds.
But being fast is not enough, the ideal is to be early. Building a relationship and starting diligence should start at least six months before startups think of raising. So how do you do that? The best way is by building a network to get the inside scoop ahead of time. Though that takes years. Another way is by reaching out to companies a year or so after their last fundraise. The challenge then becomes prioritizing which to reach out to.
So there is a resurgence of using machine learning to identify high potential startups. A scoring mechanism of sorts. Venture databases like Crunchbase and CB Insights have been selling their scoring product for years now. SignalFire, one of the first firms to build a franchise around this back 2015, spends $10M a year on its data platform. Hone Capital, Rocketship.vc, and 645 ventures are others that are public about data programs.
As an ex-data scientist, I had to experiment with it as well. So I pulled Crunchbase data, ran a regression, and stack ranked factors that drive the probability of raising a series A. Surprise, surprise (not). Getting seeded by Seqouia is the largest factor. Though this is an actionable signal, it is not a good one. The goal is to identify startups that have unicorn potential, not just series A potential. And why wouldn't Seqouia continue to fund it all the way to an IPO as well?
As I contemplated building an algorithmic system from scratch, I got counseled that it will take a team of engineers and data scientists a few years and a few million dollars to build a proper one. First, I had to identify what signals are the useful. This is a continuous cycle of data wrangling, building models, and appending new datasets. The next step is building the data pipeline and building a UX workflow on top of it. A daunting project for myself.
Luckily, Ali Tamaseb published his book Super Founders where he discusses the factors that set startups that become unicorns apart from the rest.
There are researchers who conducted similar studies, but most of them are useless. Their general conclusion was that raising the most amount of money is the largest determinant of becoming a unicorn. There are two issues with this. First, it is a multicollinearity problem. Valuation and money raised are highly correlated. Second, it isn't actionable. Early-stage investors cannot peer into the future and know which startups will raise a lot of money. Unlike those studies, Ali only looked at data available when startups were at their earliest stages, which is mostly just about the founders.
Here are the books' findings. What would have been great is a relative ranking of these factors. I'm sure Ali did the analysis. But that'll be his secret sauce.
Most fit the common patterns that venture investors look for. But two surprising findings, at least to me, are that founders' industry experience and having technical founders are not as important. I'm hypothesizing that it is because the world is more open than in the past. Industry knowledge is accessible. Venture networks, LinkedIn, etc. The abundance of talented engineers (though ironically, always in shortage) and cloud computing also made building tech businesses easier.
Building / buying the system
Now that I know what the signals are, I had to build the system. Luckily, again, I met a data startup Harmonic.ai that extracts the exact signals that Ali highlighted and presents them in a neat dashboard. I had the suspicion they just productized the book. But surprisingly the founders haven't heard of Super Founders until I mentioned it to them.
While it hasn't been long since I've used the system, I've started overthinking the meta-second order effects of these systems.
Exploiting patterns is how investors made money. Hedge funds and public market investors managers look for undiscovered trading patterns. Venture investors build a mental framework based on successful companies. The difference though is that public market trading is an exchange between two shareholders. The exploitation of a pattern means that it just gets arbitraged away. Venture investing is financing founders who want to build innovative products. Exploiting a pattern can mean looking for the same type of founders. The often male Stanford, Google, rocket ship startup alumni. I know that venture is rarely just about data and this thought exercise might just be mental masturbation. But it is a good reminder, at least for myself, that decision making should reflect personal values. There are some signals to prioritize to reflect what you believe in and other signals to just ignore.