Google, ChatGPT, and Disruption Theory

Thanks for reading! If you enjoyed this, share it with a friend or colleague, or you can follow me on Twitter @SashaKaletsky.

This week Packy McCormack dropped a wonderfully-written and fascinating piece on his Substack (yes, he is not yet using beehiiv; give him time): The Unbearable Heaviness of Not Being Positioned. It scratched many of my itches: AI, check; business theory, check; disruptive innovation, check. It was an intelligent, relevant and well-researched piece, as usual.

I felt conflicted as I read the piece: at the same time as being inspired by his use of one of my favourite business frameworks, I just simply couldn’t agree with it.

So today I’m doing something different to my usual articles: I’m *counter-positioning* this post. Packy is somebody whose writing and clarity of thought I respect a lot, so I take up David’s slingshot against his mighty Goliath sword, with a healthy dose of humility and respect.

His article is about Google, and whether it is being disrupted by LLMs. He argues it isn’t, and I think it is. I’ll explain why, finish with some overall thoughts on what LLMs mean for Google, and how they might respond. Let’s dive in and get socratic.

Disruptive innovation and Clay Christensen

I first read Clay Christensen’s book The Innovator’s Dilemma on the Trans-Siberian Express from Beijing to Ulan Bator in 2012. I am not totally sure why I read it then, I think I had probably Googled “best business books” and Google gave me a list of links from which I merely selected, as any pre-ChatGPT neanderthal would have done over ten years ago. I think the other book I read that trip was Atlas Shrugged so I’m pretty sure any brain cells gained from Christensen’s book were pretty much eliminated by that.

A few years later I read Christensen’s The Innovator’s Solution, which I actually think is a better book. I fast became a Christensen fanboy, and started thinking about how disruptive innovation theory might apply to pretty much everything in my life. On my first day at Harvard Business School in 2017 I met Christensen himself, and I chose to study the course he developed: Building and Sustaining a Successful Enterprise. I ultimately did an extended study on Christensen’s work and how to expand the course. Admittedly this was partly to use up class credits so I could enjoy 5-day weekends and spend my last semester, as we say in England, “dossing off” (Americans: that’s not what it sounds like, look it up).

So what did I learn about disruptive innovation?

I strongly recommend anyone to read Clay Christensen’s book The Innovator’s Solution to get a full picture, but the next best option is this one from HBR (cowritten by my professor). From Packy’s piece:

In short: the innovator’s dilemma occurs where an incumbent is happily serving their best customers with high-performing and profitable products, and then a “disruptor” comes along, offering lower end or new customers with products that are less profitable, or inconsistent with the incumbent’s existing offering. So at first the incumbent mostly ignores (or even ridicules) the disruption, allowing the disruptor to improve its product until the incumbent’s existing customers become addressable, until the incumbent is reduced to nothing.

Examples Christensen pull from include such exciting examples as mini steel mills, increasingly small disk drives, Walmart vs. department stores, and digital cameras vs. film. Sadly, Christensen died in 2020, so he will never be able to weigh in on the discussion.

How does this apply to LLMs and Google?

The argument is that ChatGPT is not disruptive to Google is that AI is a sustaining innovation: Google requires multiple clicks to find just any chicken noodle soup recipe, whereas with ChatGPT it’s all in one interface and I can specifically ask for it to be overcooked and extra salty, just like my beloved late grandmother’s. As we established, disruption theory states that sustaining (ie non-disruptive) innovations are one that serve existing customers better. This AI stuff sounds much better for users, so it must not be disruptive, right? Right?? Wrong, I think.

The problem with this analysis is that it treats Google’s customers as its users, whereas in my view its customers are advertisers. As has been clichéd to oblivion: we users are merely the product.

And what do LLMs and ChatGPT mean for advertisers? For now at least, they dramatically underserve them. Google’s business model is to give users the links they want, and make link-owners compete against each other financially for clicks. ChatGPT’s model is to give users the information they want (and eventually to perform the tasks they want). And information is a lot harder to sell ads against than clicks.

And to make matters worse, ChatGPT’s monetisation model is subscription and via partnerships; they are under no pressure to integrate ads into their platform. Users are being trained, in real time, to expect no ads in their LLM outputs. Google did $278 billion in 2022 revenue (and CEO Sundar Pichai made $226m), and the vast majority is of this party is funded by ad sales; they would need a hell of a lot of $20 monthly user subscriptions to make a dent.

Above are some of the most expensive keywords on Google Search. Pretty much all of them could be accessed by ChatGPT; a plugin could even find the contact information of the “best motorcycle accident lawyer” and automatically contact them. No advertisers needed.

To me (and, in my view, the late Clay Christensen), on the advertiser side this is a classic case of disruptive innovation.

What does this mean for Google vs. OpenAI?

Does it really matter whether LLMs are disruptive to Google? In one sense, no. Who cares whether a theory from the nineties applies to Google or not? Nobody at Google or OpenAI is going to be dialling the Business Academia Ivory Tower Hotline.

In another sense, whether Google are being disrupted *does* matter. The point of theories is to use them to make predictions. In this case: if Google is being disrupted, the theory would predict that they are going to struggle to respond. This is where I stand: my view is that it is going to be extremely difficult for Google to build an LLM that competes with ChatGPT.

So what can they do? Christensen’s recommendation would probably be for them to get a separate team to compete with ChatGPT (and Google) and build the best damn LLM they can. That’s sort of what they’ve done with Bard. But even that is difficult: corporate inertia, conflict with ad sales interests, and shareholder pressure (AI engineers are more expensive than any number of free massages that Google have rolled back). The other thing they’re doing is integrating AI into search results themselves. But they can only do this so much without conflicting with their ads product: you can be pretty sure they aren’t going to give you the email address and dial you directly into the best accident motorcycle in your town.

This is the thing about the innovator’s dilemma: it’s a dilemma. Any solution will include sacrifice. Packy and others have made the point very well that in the long-term, integrating AI will improve their user experience. But I’d argue that’s not enough: Google’s urge to satisfy its advertisers could prove overpowering.

So the real question that will govern Google’s strategy for the next decade: who do they really treat as customers: users or advertisers? Unfortunately for their long-term future, I am betting that it will be the needs of the latter that dominate over the wants of the former.