How To Compete With AI Flywheels

Much has been written over the past few years about the “flywheel” nature of AI. More data means a better product, which means more users and more data, and so on and so on. Flywheels of any kind are tough to compete against because once they are set in motion, they just go and go. Imagine, for example, trying to compete with the Netflix recommendation algorithm. How would you ever build something better given the flywheel they have for making that better?

I believe there are some opportunities. Some may not be available at the moment for whatever industry you are in but, I want to highlight them here briefly so you can keep your eyes out for them if they arise.

1. Offline or unique data sets. One of the things no online algorithms know about me is that I love to read about accounting fraud. I had a professor in grad school who taught accounting this way — every time he taught us about an item on the Income Statement, he gave examples of how people game that item. It was fascinating to me because accounting, from the outside, seems like a very exact science. It’s not. I don’t go around the web reading accounting articles in general, so usually I spot these articles in a paper version of the WSJ, or maybe the Economist. If you had access to data that mattered about a decision, that wasn’t easily available online, you could possibly use that to compete against a large online AI company.

For example — I wonder if an ioT world will give Amazon a run for its money on ecommerce. Could there be a future where the in-store shopping experience is better and more personalized than the online shopping experience because of sensor data measuring new things about your shopping experience that weren’t being measured before? I don’t know. Perhaps the online data is waaaay more valuable. Or maybe ioT is a chance to compete. (Maybe this is why Amazon is opening physical stores). Keep an eye out for new data sets.

2. Market Resegmentation with new algorithms. “Recommendations” in general are done by a handful of well know algorithms like collaborative filtering, but, sometimes if you look closely, there may be new opportunities to pull off a piece of the market where an algorithm has been over-applied and something better exists. I only realized this when Katerina Axelsson from Tastry approached me at a conference to talk about her AI work on analytical chemistry and taste recommendations, and how, for sensory items, she could beat collaborative filtering. (Disclosure: I invested) It turns out just because you and I like our coffee the same way doesn’t mean we like all our other drinks the same. So a better recommendation engine will get at my underlying sensory interactions rather than higher order articulated preferences. I bet there are many similar opportunities out there being overlooked.

3. Small data. Strategy says that you should double down on areas where you are strong and have assets that are difficult to match. It’s no surprise then that the work coming out of Google, Facebook, Amazon, and the like is mostly ML/AI focused on big data. They have unmatchable data sets. But there is no reason to believe that small data AI isn’t an opportunity. While many people are hoping that blockchain will allow customers to control their own data, and thus put a dent in the tech monopolies, it’s possible that simply solving some of the small data AI problems would do the same by removing the big data scale effects from the market. I’m hopeful that AI hardware will have a hand in some of this.

When you look at AI today, flywheels are winning, but keep your eyes open for opportunities like those mentioned here that will let you go smash someone else’s flywheel, or make it less relevant for some part of their market. And if you are building a company around one of these ideas that can beat an existing AI flywheel, drop me a note. I’d love to invest.