The Commoditization of Machine Learning
Machine Learning (ML), in all its various forms, is as hot as any sector in the startup world these days, both in terms of capital invested and M&A. This week our issue features a great piece from CB Insights on all of the industries being disrupted by deep learning.
Part of the explosion in ML-based startups is due to the commoditization of ML technologies, thanks in part to open source algorithms, cheap cloud computing, and widely available training data sets. We are currently at the point on the curve where the cost of implementing ML has come down enough that startups can quickly throw together a ML-based application leveraging AWS’ ML engine (or other comparable service). However, the commoditization will not stop here. Large tech companies, and academics, will continue to push the boundaries of AI forward, and share new algorithms and approaches with the open source community, because after all it’s the data that is really the differentiator for these large players.
As large tech companies continue to open source their innovations it will continue making it easier for startups to “plug AI into their products” (as Kevin Kelly might say). However, that also means that it will become nearly impossible for startups to build defensible business models by just offering the ML-powered version of previously ‘dumb’ applications, because the barriers to delivering ML-powered solutions will be so low. I realize that I have oversimplified this argument, and there certainly will be some barriers to ML access, a scarcity of talent to implement it, and unique ML approaches for specific applications, but directionally speaking this is where the space is heading.
The big challenge for investors when considering backing an ML-based company today is, they either have to: a) bank on a quick exit before the technology becomes commoditized, Or b) have confidence that the team can build a defensible product capable of standing up to the large tech players and sustaining its value prop, even as better and cheaper AI permeates everything around them.
One of the trends I am seeing is that investors in this second camp, are looking for companies that are not only offering ML-as-a-service, but are developing a much more sophisticated end-to-end solution, involving a deeper level of engagement with their customers. While this model is by nature much more involved and difficult to build, it leads to higher switching costs for customers and also enables the startup to own some of the critical touch points, such as user feedback. This relationship gives the startup a defensive edge that other incumbents cannot easily replicate. I continue to be fascinated by opportunities in the ML space and love to see what crazy new applications entrepreneurs can find for AI. However I think that over the next two to three years the landscape will change dramatically and the exit opportunities will shift to favor those companies providing unique solutions that large payers cannot easily replace or commoditize.
-Mike Droesch, Founding Editor