Machine learning is fundamentally changing the way we develop software. Advances in algorithms, data set availability, and GPU hardware over the last few years have a compounding effect, and the capabilities of ML systems are improving exponentially. Hence ML is a deeply disruptive technology that will affect every industry.
This creates huge opportunities for a new breed of companies — and a huge premium for figuring out the blueprint for what makes them successful early on. There has been a lot of discussion about who will be able to capture the value AI/ML systems create. Will the big players like Amazon, Facebook, or Google monopolize AI? Will AI do away with the economy all together by driving the marginal costs of everything to zero?
We think that there will be plenty of opportunities for startups to create huge, defensible businesses based on AI/ML.
Our hypothesis is that a good way to achieve this is via data network effects:
- Build a product that is sticky, addictive, and provides value even with basic AI/ML
- Include a feedback loop so that usage improves your models
- Use this to start a virtuous flywheel effect that lets you outrun the competition
Google Search is a great example of this: more users => more click data => better ranking => more users. Makes it virtually impossible for anyone else to catch up.
Additionally, we think it is necessary to tackle a problem which is sufficiently routine, narrow, and has a clear enough target function to make current ML approaches applicable. And to have the right mix of skills on your team to build both a sticky product and train the best in class models for your domain.
The graphic below summarizes our investment thesis on machine learning
Let us know what you think. And if you are working on a company that fits this thesis, we would love to hear from you of course!