How To Get A Piece Of The Boom In Machine Learning Start-Ups
Machine Learning is hot. Dozens of machine learning start-ups have been acquired by big companies over the last five years, including two this week: Apple is buying Turi Inc., and Intel is buying Nervana Systems. Investors are eager to put money into credible ML businesses.
Machine Learning (ML) describes systems that can learn without being programmed for a specific task: they find meaningful patterns in data and use those patterns to draw conclusions that lead to useful results.
What is happening here? As a technology, machine learning is breaking through from the lab to the real world. It can be used to solve many specific problems, e.g., a Boston-based company is using machine learning to improve by orders of magnitude (multiples of ten) the ability of computers to solve a difficult, large-scale data management problem [details confidential] that crops up all over the world of commerce and is especially acute when companies consolidate computer systems or merge. And, many robotic systems incorporate machine learning techniques.
Machine learning also enhances the effectiveness of many of the big technology platforms employed by the major technology companies. You may have noticed that voice recognition has improved substantially in the last few years. Voice recognition is a key part of Apple’s platform (Siri), Google’s (voice search, Google Voice message transcription), etc. ML can also be used to improve the algorithms that determine which ads get displayed or what appears in your Facebook feed. So it is highly strategic for the big tech platform companies.
Hence, you see many ML companies starting to focus on many different problems, a huge scramble for ML talent, and many early acquisitions by the big tech companies, some of them said to be at impressive (if unconfirmed) valuations.
What does this mean for the strategies entrepreneurs can use to participate in the ML boom? First, competition in ML currently is quite focused on talent. The number of experienced and qualified people is limited. So, your ability to pull together a highly qualified technical team is critical. An ML entrepreneur told me that it’s almost impossible to start an ML company in Silicon Valley now because of the intense competition for people with the big internet companies. Locations like Boston or Pittsburgh are a better bet: there is a strong supply of qualified people coming from the universities, and less hiring competition. Europe has potential too: some great universities and much lower level of start-up activity than the U.S. DeepMind, which was acquired by Google and produced the system that beat a human Go champion, came out of Cambridge University in the U.K.
The exits have typically come early, because the major companies are looking for talent and technology. Entrepreneurs need to decide if they want to play for an early exit, or try to build a substantial business. In either case you want the best possible talent. If you expect an early exit, then keep pushing the product ahead and demonstrate what it can do, but avoid taking a big slug of money for commercial scale-up. In a modest exit that money will come back off the top with a return, reducing the value of the common shares significantly. It will be better to raise modest amounts of money from angels and small funds, until it becomes clear you want to strike for commercial scale-up. This goes against the “take money when it’s offered” rule, but in the ML market right now entrepreneurs usually have multiple financing options.
If you see the chance to build a substantial business, you need to think about the risk of getting crushed by the big tech platforms. Look carefully at what they are doing and decide if your chosen area of focus is something they will need to control, or if you can co-exist with them. Companies adjacent to Microsoft had to make similar judgments in the PC era. The more distinct your domain knowledge is, the better the chance they will choose to co-exist, or buy you at a price that rewards the value of your business as well as your talent and technology. So if you can find a big important problem with deep, specific domain knowledge that is probably not on the must-control roadmap for GOOG, FB, APPL, IBM, or MSFT, you have an opportunity to build a big standalone company.
Amazon web services (AWS) is an interesting platform for ML companies. A company I know is selling an ML based web service on the Amazon platform with notable success: customers send in data sets, and the ML web service finds the patterns in the data and returns them as a response. Start-ups have been a key customer base for AWS from its early days, and it continues to be a friendly platform for them.
ML shows strong evidence of being a new, new thing. Investors’ and acquirers’ check books are open and entrepreneurs are excited. I’m hopeful that, with the right approach, entrepreneurs can build some great products and great companies here and, with their investors, make some money. And, ML creates new areas of opportunity in a tech world that has come to be dominated by a small set of topics and companies in recent years.
Originally published at www.forbes.com on August 15, 2016.