So I never really divulged the AI project that I was working on several months ago, and why or how we got outcompeted in the early innings. And the reason is because there is still some space left to maneuver there. We did release an early stage product demo, x.GPT, which later become Modler, but these products were always stepping stones (features) leading up to the main event (still secret, sorry).
Last week I wrote more about where the AI value return is going to be, but I don’t want readers to conflate “most” of the value with “all” of the value. One of the most destructive perspectives a person can have is that of a zero sum mindset. In a zero sum game, the size of the pie is static or fixed, meaning that you can only trade existing economic pieces back and forth. There is no growth. Oligopolies become deadly in this game, sucking up all resources leaving none for anyone else. In reality, when the right incentives are in place, the size of the pie can increase, creating a “positive sum” game. This means that oligopolies can actually be good. Even though they increase the size of their own value at a faster pace than everyone else (which effectually increases the inequality gap), in doing so they increase overall productivity (and opportunities) for everyone. This is the lens with which you can view the AI space. Are you better off throwing all your money into GAMMA stocks (the oligopolies) and relaxing for the next decade, or grinding every single day trying to find those 100x and 1000x opportunities? The former might be more prudent, but the latter is obviously way more fun. One thing to note about wading through the current startup market though, is that (like generative-AI content), there is going to be a commoditization effect leading to a ton of absolute garbage.
I was talking with one of my business partners this week about our outstanding AI project, and we both agreed that there is still massive potential, but when the full version of GPT4 (with all the code interpreter and file uploader bells and whistles) is released to the general public, it is going to commodify many software services and features. As I wrote last week, my Front End Theory still holds, but if you have 200 chatbot applications all doing the same thing, the price a consumer is willing to pay will quickly be competed to zero. This means there will eventually be no money in the traditional subscription model, and another method of monetization is required.
Acquiring attention spans to sell advertising has been the obvious way to offer a free product to users over the past 15-20 years or so, but there are other modern ways to make money when the product is free. There is the “freemium” way, which provides a bare bones application that services the majority of the users for free, and then charges a premium for top customers to enjoy bonus features. That will definitely get some play in the upcoming years (already seeing it with ChatGPT Plus), but with such rapid commodification, and rabid competition, it doesn’t sound like the best bet to me. Another popular modern method is to simply sell your user’s data. Robinhood monetizes this effectively through a method called “Payment for Order Flow” or PFOF.
Contrary to popular belief, Facebook’s value in its advertising model is not inherently in the user’s data. Facebook (I sometimes refuse to call them Meta when discussing their core features) is a marketplace that matches up buyers and sellers. The user’s data is imperative within the walled garden, but only in so far as it contributes to the web of demographics, interests, and social connections that grease the machine. Outside of the walled garden the data becomes exponentially less valuable. Sure, you can take a corpus of demographic data to craft insights, but what is the best thing to do with those insights? Pump them right back into the machine to target more customers and sell more ads (or influence elections if you are feeling particularly spicy). A single user’s Facebook data has exactly zero worth in the open market, no matter how many ads they click or convert on.
Conspiracy theorist side note: all of this privacy stuff is just marketing by Apple to take a leg out of Meta, paired with them sucking up to the government as Big Gov rails against Big Tech for making trillions of dollars by “monetizing our data”. Go ahead, opt out of being tracked, you will still see ads, they will just be less relevant, and your app experience will likely suffer. Like I wrote in one of my earliest articles, consumption is not going away, and matching buyers with sellers is actually good for the economy. No one is forcing you to buy anything. If you think Facebook is evil, you probably lack the mental constitution to think for yourself and default into seeking the approval of others. Eek, sorry. (Not sorry).
Robinhood’s PFOF model is the opposite. Robinhood is able to allow users to execute stock trades commission free by selling that trade activity in real time. The actual “customers” are institutional investors that get to front-run what the retail investor is doing by injecting themselves in front of that user’s trade, buying the stock and selling it to them, making money off the spread, all within the fraction of a second. This is why I have said that retail “trading” has no shot against institutions, and that if you want to play the market as an individual it has to be via “investing”. Anyways, a topic for another day. The value of the PFOF model is that it takes an individual’s data (not an aggregate web of trends), and sells it directly to the highest bidder in real time, outside of the walled garden. This data is extremely valuable to outside observers because it is transactional, definitive, and actionable. A single user’s Robinhood data has value to many participants in the open market, commensurate with precisely the bid-ask spread of each trade said user makes.
Most experts in the space recognize that AI chatbots are are going to be a new platform paradigm shift the way that mobile apps or GUIs were in decades past. This is partly due to the fact that it is always easiest to predict the future that is right in front of our faces, but also because these AIs are at the core just language models, which necessitate a certain form of interaction (language). The future being predicted is that every person and organization will have an AI chatbot that they can query for information, and connect to other AI agents to go out and perform actions in the digital world on their behalf. Open source code will eventually commodify thi basic chatbot feature. Simply ETL your data into an API and you will be off and running. Connecting to other AI agents to perform actions provides an interesting opportunity, however.
Let’s say that search engines such as Google and Bing go away, as many are predicting, because the interface in which we interact with the internet is via one of these AI chatbots. This has the potential to unlock a tremendous amount of commerce, even more than the biggest unlock of all time, digital marketing. Digital marketing, in all its dominance, still struggles as humans try to optimize it, constantly having to balance social proof, value, pricing, intent, against the constant battle of data privacy and demographic identification. At the end of it all, of course, is a conversion event. AI chatbots and their connected agents can cut through a lot of this noise surrounding the customer journey, with intimate knowledge about their owner’s needs and desires. Perhaps at some point they know what we need more than we know ourselves. This knowledge can apply to B2B, DTC, or any configuration of polyamorous ways that may currently be in vogue. The end result is the same. The reduction in friction yields an increase in product-market fit, and thus commerce.
Imagine an AI service where someone can not only front run something before you buy it, but also convince you to buy it, or even just buy it autonomously with pre-approval knowing you need it. A real time bid ask spread could exist for essentially any product or service. The AI can identify how much you would be willing to pay, and sell that information to a third party who can buy and sell you that product, netting a spread just like the institutions who profit off PFOF. Right now, resellers for things like CPG, or even utilities, exist because they buy products at wholesale prices, and (due to their proximity to the customer) can mark up to MSRP. Much of this intermediary value-add has been abstracted away over the digital age via companies like Amazon and Walmart, bringing manufacturers and consumers closer together. Thus, the value-add, or monetization method of your commodified, open-source, AI chatbot product (or feature) is in its ability to connect the user to the economy. Need users? Offer the product for free. Need money? Sell that user data to the highest bidder. Make no mistake, AI products that can query a walled garden will be incredibly valuable to a consumer, but your landed cost will be whatever electricity and compute bill gets passed through to you. AI products that connect beyond the walled garden and interact with a marketplace will be the ones to build or invest in.
Don’t conflate the analogy of user data with the market cap and overall business models of Meta and Robinhood respectively, though. Meta’s market cap value is that of insular network effect, Robinhood’s of exogenous intent.