Sign in Subscribe
8 min read

Google and the Innovator's Dilemma

This article started as an excuse to recap the Innovator's Dilemma for myself. If you're familiar with the book and the theory, you can probably skip at least the first half of this article.

The Innovator's Dilemma

The Innovator's Dilemma is a theory devised by Clayton Christensen and is detailed in his book of the same name. Anyone interested in business strategy, particularly in the tech world, should read that book and anything else he's written. Here, I can only recap the high-level details of the theory. Everything in this section of the article is taken from that book.

Every business has products or services that compete along one or more dimensions. Take shipping as a random example. It competes based on price per kg, speed, or a combination, depending on what's being shipped. As another example, computer processors compete on absolute speed, and those destined for laptops also compete on power efficiency.

All businesses are then embedded in value chains that value those dimensions. So, for example, a global container shipping company is part of a value chain that seeks the lowest possible price per kg of goods moved without excessive concerns for speed. An air freight company, in contrast, works with businesses where speed of delivery matters more than cost. That value chain appreciates improvements along the dimensions they care about (e.g. if container shipping gets cheaper) but cares little about other improvements.

On top of that, every business is potentially subject to innovations in the space in which they operate. These are any technological improvements that can impact the quality of the product or service being delivered. Those innovations can be split into two groups:

  • Sustaining: These innovations allow a product or service to improve along the dimensions they're already competing in and their value chain desires. Any business is strongly incentivised by the value chains it operates in to adopt these innovations as quickly as possible, and they generally do so.
  • Disruptive: these innovations result in a product or service that is generally less good than what already exists along the competitive dimensions already established but better along some new dimension. The problem is that the value chain the business is embedded in doesn't value that new dimension. So, there is no incentive (in fact,) the lower performance means an active disincentive) to make use of the innovation.

Returning to the processor example, x86 has been the dominant architecture for much of computing history and competed based on raw performance. The ARM architecture was a disruptive innovation which was less powerful overall but much more power efficient. The existing x86 value chains didn't care; they primarily made servers and desktops that were plugged in. The new value chain of mobile devices did care, though, because they had to run off batteries, and the longer those lasted, the happier consumers were. That's where ARM won initially.

The final important consideration is that technology generally improves faster than the market needs it to. ARM chips have been improving over time, and their performance is good enough for servers and desktops while still being much more power efficient. In principle, that means that they could start to disrupt x86. The big hyperscalers, in particular, are now designing their own ARM processors to run in servers because the performance is at least good enough, and a drop in their electricity bill is significant. There are other reasons, to do with backwards compatibility and the moat around x86, that this transition has taken a long time and will be slow even now. But ARM processors found a value chain they could dominate, improved over time, and now threaten to disrupt a value chain they were previously unsuitable for.

What makes this a 'dilemma' though? Well, it's the fact that the business being disrupted, the one in the old value chain, is essentially forced into this situation by a set of rational decisions. Of course, they won't be able to invest lots of money in a new technology that is worse at what their value chain cares about. Their existing customers, around which they have built a successful business, demand improvements on the existing dimensions. Allocating resources for the benefits of a different (and in some cases, currently non-existent) value chain risks falling behind on the dimensions that have built the business, losing those customers, and not having the money to do anything. It's generally impossible for a company to allocate significant resources towards a disruptive innovation until it's too late without jumping through a bunch of hoops Christensen goes through in his book.

How does this relate to Google?

There's a strong argument that Google, by which I mean their ad-driven search business (i.e. the part that counts), is being disrupted by AI. To think this through, let's consider the value chain they're a part of and what that value chain values.

flowchart LR Users---|Want to find stuff\non the web quickly|id1((Google)) id1((Google))---|Want users to \nvisit their site|id2[Organic sites] id1((Google))---|Want users to click\non their ad|Advertisers

While simplistic, this captures the essential parts of the chain. Google is an aggregator. If you don't know what that is, I strongly recommend reading Ben' Thompson's Stratechery, particularly these articles. You'll need to subscribe for at least some of them but it's easily worth the money, and I don't get anything for saying that.

In brief, an aggregator is a business that succeeds not by controlling supply (as old-school monopolies did) but by facilitating discovery in an abundant space, and thereby controlling user demand. In the diagram above, users go to Google when they want to find something online. Businesses, therefore, have no choice but to be on Google to be found by potential users.

On the right-hand side, I've separated advertisers and organic sites. The advertisers are likely also sites, and they do, in fact, want visits just like organic sites; the only difference is whether they've paid to be on the Google results page. I want to talk about the impact of GenAI on organic sites in the future, but this article will focus on advertisers. Google has a lot of power over advertisers, but that power comes from the aggregation of users.

What do advertisers value?

In this chain, four things matter to advertisers:

  • The number of unique users of the platform
  • The frequency of user visits
  • The duration (in terms of both time and the number of clicks) of user visits
  • The amount of data available about a user

The first three are obvious:

  • 1 billion users give access to more potential customers than 100 million
  • If users access the service multiple times a day, there are many more opportunities to sell them stuff than on something they use once a month.
  • Users who click through multiple links and explore are more likely to see and respond to an ad than users who immediately find what they want and leave. However, it's worth noting that there's a tension here with user experience: users typically want to be done as quickly as possible.

The fourth point, data, is needed for targeted advertising.

You can roughly break advertising down into two groups: contextual and targeted. Contextual advertising uses just the context they're currently in and no outside information. So, for example, if a user searches for 'holidays in Greece', you might show them flights and hotels in Greece, some summer clothes and a blog about travelling in Greece. In contrast, when using targeted advertising, you know other things about them. Again, simplifying, you might know they are young and lack disposable income. You could show them hostels to stay in, cheap backpacks and a blog about visiting Greece on a budget.

Targeted advertising is way more effective and benefits everyone. Advertisers now show their products to a group of people who actually might want to buy them, and users only see things relevant to them in the ads. Obviously, concerns about data and privacy are valid, but I won't address them here; the point is that advertisers value the ability to segment users.

What do users value?

Users go to Google to find something online and do so as quickly as possible. There are many different ways to break down what people search for, and I've realised I will have to do a separate post to cover all the details. Let's consider two possible ways to break down queries for now. Some things will benefit from personalised results (e.g. "summer clothing"), and some won't (e.g. "temperature in Athens"). For some of them, a 'good enough' answer will do (e.g. "Greek dishes for vegetarians"), and some need to be the best possible answer for you, even if there is no 'right' answer (e.g. "hotels in Athens").

Where is Google being disrupted?

Google has all the user data, and most AIs (like ChatGPT,) for example) have very little. Anything that benefits from personalised results goes to Google. I would also argue that anything where you need the best possible answer still goes to Google's approach of showing me many options and making me pick. No matter what Silicon Valley might tell themselves, the days of me asking an AI to find and book a hotel for me, without me scanning through a list of options like I do today anyway, are a long way off. I don't know what criteria I use when booking a hotel beyond some basic filters around price, rating and location; once I get down to a short enough list, it's as much about gut feeling as anything. When I'm going on holiday, I care about finding the best possible hotel, given my constraints and preferences, and a list of options with sensible filtering criteria is the best way to do that.

But what about where a 'good enough' answer will do, and that doesn't depend much on personal information? My example above was for a list of vegetarian Greek dishes; even if you argue with my example, you can see my point. With ChatGPT as the best current example, AI does this way better than Google ever did. I can get a first attempt in less than a minute and refine it with additional prompts. The first few times I did this, I must admit, I then went back to Google to make sure nothing obvious had been missed. Now that it's evident that ChatGPT does a good enough job, it's the only place I go for planning, brainstorming and other forms of idea generation.

What does this mean for Google?

Interestingly, from a user perspective, there's a strong argument that this isn't disruptive. Users have always gone to Google to get the information they need as quickly as possible, and now there's a technology that allows them to get a subset of that information even more quickly. There's no doubt that Google's behind and executing clumsily to catch up, but it's less clear that they've been disrupted (in the Innovator's Dilemma sense) on the user side.

On the advertiser's side, though, one of the important dimensions was how long users spent searching and clicking for things. Showing users an answer immediately so they can move on more quickly is worse for advertisers. They will be seen and clicked on less, and Google can extract less money from them. AI has exacerbated the tension that has always existed between users of Google and advertisers; their goals have always been slightly at odds, and AI has given more power to the user side. This feels like the Innovator's Dilemma for aggregators, and it's very similar but subtly different from the original.

GenAI is better than an old-fashioned Google search for a subset of activities detailed above. Still, Google rightly recognises that if they give GenAI that foothold, then over time, it will improve and become a threat in other areas. To avoid this, Google has crammed AI-generated results into everything, improving that subset of search results and making all the others worse.

It's clear Google needs to be in GenAI, but they had every advantage and have still ended up behind and on the back foot because it's disruptive to their business model. However, introducing GenAI results for all users and all types of search at the same time feels like a panic response, and a more targeted introduction (which I have little doubt Google is technically capable of) might have allowed them to reduce the threat from GenAI without taking a hit on the rest of their business.