Machine Learning to predict the startup success?

, , 2 comments

When I first heard about Machine learning I was really pissed off because of the below three reasons

  1. Machine learning, F*** yet another technology to make life complicated.
  2. I should again update my small data base (brain) with this new technology and should override the technologies this might erase. Hectic job L…!!!
  3. Can someone write program to keep human mind updated automatically with the new technologies inventing all the time. I hope this happens soon.
With this notion I started reading about machine learning and to my surprise it’s actually fascinating and interesting than I ever thought. Machine learning is fascinating because you cannot achieve the capabilities and results seen by machine learning methods with any other methods. The most fascinating about machine learning is you make programs to learn from the data, examples and experiences and that programs make programs to address different problems.

Curious to know more about machine learning I started researching about the usage of machine learning in different fields. I found that machine learning has given us self-driving cars, practical speech recognition, effective web search, and a vastly improved understanding of the human genome. Machine learning is so pervasive today that you probably use it dozens of times a day without knowing it.

After all this research the next question that struck me was how can I make money with machine learning???

I was reading an article in FORBES which points that 80-90% of the startups fail every year. This struck me with an idea on why cannot we use machine learning to predict the success of the startups??

I guess with some effort it is possible to gather data from these ventures and also from the other successful ventures which could actually be the data set for the machine learning methods. Data about thousands of companies can also be collected from online sources such as CrunchBase etc.

At this point of time I have no idea about the complexity of addressing this challenge and the number of variables that need to be considered to get more accurate results but I am sure machine learning can definitely be useful in this regard to determine the exit likelihood of private companies and explain why they might succeed or fail by highlighting their strengths and weaknesses. This analysis can also be helpful to suggest ways to improve the distressed startups.

The next question that bothered me was, do any venture capitalists use machine learning to predict startup success?

For Venture Capitalists, predicting, measuring and evaluating the success of the startups they invest in is risky business. As per my knowledge (I admit my knowledge is limited in this area), Very few VCs actually use software with machine learning techniques to predict success. I believe the reasons as below

  1. VC’s sample is too small compared to the market, so it's difficult to allocate money to build a piece of software to process small sample data.
  2. Even if VC’s are planning to expand their databases with other VC portfolios, they wouldn't have enough accurate data and insights about those companies.
  3. Humans change plans. People is the most invariable factor in every single company. It's also the most important one because people have an impact on everything and everything can have an impact on the people. I believe investors almost always invest in the right combination of management team and product and not just only product. In this context no software can accurately predict the human behavior and this is a major bottleneck for creating such software.
VC’s at the moment use tools like Mattermark in order to detect and track startups, sectors and market trends and try to bid on the winner of each category.

I believe a software with machine learning technique would address all above problems and provide VC’s with more accurate results than they would get from other sources.

As per my knowledge very less work is being done in this area, I suppose only one startup (Trenify.io) is actually working on implementing this but I suppose their accuracy is also based on certain limitations/assumptions on human behavior.

I believe Machine learning to predict the startup success could be one area for people planning to launch new ventures.

I would love to see people commenting to this post and discussing if this is a viable option for a startup?? Will this startup idea be successful before it predicts the future of other startups???


2 comments:

  1. Hi,
    I work as a data scientist at a VC (Balderton Capital) and I have been working on the kind of things you mention for a while. I am also a machine learning guy. My thoughts:

    To some degree, you can make use of data about startups, and there are many startups that try to do this data collection/aggregation already, most notable examples are MatterMark and CBInsights. MatterMark had a famous single score that was meant to predict startup health/success, but it didn't work very well, nor was it very well received.

    Some data is more useful than others. For example, detailed mobile app store data from places like App Annie are very useful, but also often very expensive to obtain. For stats on website visits, Alexa or SimilarWeb are very noisy as they are based on panel data. By the time the visitor figures are large enough for the sample size to make sense, you may have already missed the best opportunity to invest. Data on teams and individuals can be very useful (spotting repeat entrepreneurs, ex-Google people, PhDs, etc), but very hard to obtain: linkedIn is closing down API access making large-scale analysis very hard if not impossible.

    Fundamentally, there are limits to how much publicly available data can tell you about an early stage startup. A good example is TouchSurgery, an app that is used by surgeons. On the surface it doesn’t look like a blockbuster success, because it’s only used maybe by ~100k people or so. But, all of those people are surgeons, which is a very valuable audience. The main value of this business is capturing the engagement of this audience, and you will only understand why this is a good business, if you listen to the founders explaining things. These reasons are very vertical-specific so it’s really hard to generalise to generic approaches that would find other similar success stories.

    So overall, yes, there are some ways you can use data and a bit of machine learning for deal sourcing in VC, firms and some startups are already doing some of this, but the approach also has limitations and it is not as revolutionary as it may seem on the surface. Providing this service to VCs is also not a very good business model, because VCs is a small market, and not particularly nice customers to sell stuff to (they are frugal and good negotiators).

    ReplyDelete
    Replies
    1. Hi Ferenc Huszár,

      Thank you for dropping your thoughts on this topic.

      I would agree to some of your points:

      1. By the time Visitors figure are large enough we might have already missed the best opportunity to invest. This could probably happen and yes this has to be consider for such startups.

      2. My post is more directed towards VC's and I agree VC's is a small market but such a startup can also analyze the existing startup and determine the exit likelihood and explain why they might succeed or fail by programmatically highlighting their strengths and weaknesses and suggest the possible ways to perform better

      Or

      Such startup can be helpful to the distressed startups and analyze what is going wrong and suggests some ways to be competitive in the market by considering the data from the existing successful companies in the same sector.

      About 90% startups are not successful and so such startup I believe can be useful but to what extent it can be successful in terms of revenue is a big question.
      What are your thoughts on this?

      Thanks,
      C. Pavan Kumar

      Delete