Machine learning and AI (as it’s commonly spoken about, not in the sense of artificial general intelligence) will become a commodity. There are thousands of open-source libraries that allow development teams to tap into neural networks, deep learning, image recognition, etc. Furthermore, there are services, like AWS Rekognition and Google’s AI Platform, that mean you can get started with machine learning even without a team of data scientists and ML engineers (I’m admittedly taking some leniency with that last statement).
As compute gets better and data sets continue to balloon, software will get smarter and a host of businesses will employ ML and AI to deliver a better experience for customers. There will come a time, in fact, when nearly every new business will use basic intelligence in some capacity or another. The best of these companies will do just fine.
The irony, however, is that most of the value generated by AI and ML will not be created by these companies, but instead by the holders of data. In other words, it’ll become increasingly easier to create smart features, leading to a democratization and also commoditization of AI and ML. The unique value, then will only emerge via the businesses that are able to control data.
Just as there will be companies employing machine learning and AI to create better experiences, there will also be (and already are, in fact) startups that offer intelligent services. In the long run, these companies will struggle. Put simply, this has to do with the idea that ML models perform better when they are given a larger data set of past history to go off of. The biggest tech companies in the world have such data sets and are continually taking steps to increase their grasp of data.
Opportunities, however, still exist. First, if a company can establish partnerships with the big tech players in order to provide services that the large company is either not interested in or cannot provide, the startup can gain access to a treasure trove of data that will set it apart from the competition. Companies like Sprinklr and Datasift have shown this is a viable route.
Another way is to eschew the tech companies all together in favor of working with traditional incumbents. IBM’s Watson business, for example, has partnered with enough big, legacy companies and aggregated their data together to build differentiated and defensible products. The advertising space holds a number of examples of companies that have managed to succeed by following this strategy as well.
The most interesting possibility, however, is for a startup to generate its own unique data set and grow it fast enough to gain an advantage over incumbents. This, of course, plays into the perpetual battle of sustainment versus disruption as laid out in the Innovator’s Dilemma and is easier said than done, although Lemonade appears to have made some serious traction by following this approach. Nevertheless, the difficulty of this path means most successful AI and ML startups will have to rely on partnerships. There will be those special few, however, that manage to be exceptional.