Platform economy is not for startups anymore: you can only win with a corporate data stack
After carefully considering competitive advantage in the platform economy, I’ve come to the rather unfortunate conclusion: there are several dynamics which hinder competition, disruption and abilities of newcomers to come to market. Furthermore, the dynamics I’ll present in this blog post help governments, government agencies and few of the biggest companies of the world and make growth and even survival difficult for others.
The platform economy trinity
Platform economy is more and more about three core aspects: 1) data, 2) machine learning, and 3) a new operational model, the platform business model.
Why are these developments happening now? There are some drivers that currently move the boundaries between firms and the market. Platforms are a new compromise in between the two. I have written about these drivers elsewhere, but most importantly platforms are enhanced with a victorious loop where the quality of the platform offering leads to more users, and thus more and better data. This data enables more efficient machine learning, which in turn can be used to improve the platform offering.
This is indeed the winning model: when a platform company is able to continuosly, exponentially improve, personalise and position their offerings via machine learning, competitors have a very little change to compete.
Therefore I seek to answer to the question “what is the source of this unfair competitive advantage?” Let’s take a look at each part of the platform economy trinity individually.
Competitive advantage from platform business models
I have written about competitive advantage in platform business extensively, but I now want to focus solely in how to gain competitive advantage via a platform business model. In other words, the focus is not how to be competitive in platform business model, but how to gain advantage from platform business model). Further, it’s not the topic of this text to compare platform business models to traditional business models (I’ve done that elsewhere), but to compare two different companies both applying platform models.
As I have noted elsewhere, the competitive advantage in business models comes from the strategic choices that a company makes. These choices have either flexible outcomes, such as a price for a product, which can be easily changed, or rigid outcomes, such as the factory that produces the product, which cannot be easily changed. Competitive advantage in business models comes only from the rigid consequences, because they cannot be instantly copied. In other words, rigid consequences are consequences that imply that there is a delay in the causal connection between the choice and the consequence.
Competitive advantage in business models comes only from the rigid consequences, and platform business models have less rigid consequences than traditional business models.
Remarkably, platform business models have less rigid consequences than traditional business models. For example, in an event of a global long lasting ban on travels, a hotel chain could not get rid of their rooms in a blink of an eye. AirBnB can do this. They won’t, of course, because hotel rooms are a variable cost for them, not a fixed cost as is the case with the hotel chain.
Because there are less rigid consequences, a platform business model is always easier to copy than a traditional business model. There are many AirBnB, Ebay and Amazon clones in global the market, some more succesful than others. Nonetheless, their success is clearly not dependent on the business model, which they are able to clone completely .
Platform business models cannot be protected via a unique business model. Why, then, there aren’t hundreds of succesful platforms competing in every market?
Competitive advantage from machine learning
The Soviets did not originally learn about the Manhattan project (the US nuclear bomb test) via spies or leaks. A gifted soviet fission scientist noticed that all top level publishing regarding fission in English language scientific journals simply stopped. This alarmed him to make a report to Stalin that the Allied are constructing a fission bomb. (video)
Machine learning and artificial intelligence companies generally publish all their results in scientific journals. As long as this continues we can be rather sure that it’s not possible to get a significant competitive advantage from a unique machine learning technology. This is disappointing news for many smaller innovative companies and startups, but their belief to the contrary won’t help them. Rather, any new machine learning innovation is generally easily copied even without the specific knowledge of the technique behind the breakthrough.
Nevertheless, machine learning and artificial intelligence are a core part of the competitive advantage of the platform companies, even if it’s not the machine learning algorithms themselves that provide the competitive advantage. As explained by Andrew Ng (video), the significant difference between the previous generations of machine learning and the current paradigm of supervised deep learning is that more data there is, the better the current models get.
The previous machine learning technologies were not good enough to provide constantly improving opportunities for those companies that can increase their data trove. Nowadays when this has changed, the role of data is even more important than it was just three or five years ago.
Competitive advantage from data
So, the significance of data has counterintuitively only grown with the recent advances of machine learning technologies. Ng points out that these days many platform companies launch products only to gather data. It’s not so much a raw material — “the new oil” — as it is the source of the competitive advantage. Perhaps an oil field would be better equivalent.
This data is used for teaching the machine learning algorithms. This provides constant improvements in the products or services offered on the platform, which in turn increases number of users and thus also increases the amount of data.
The better data one has, the better machine learning capabilities one can build. The better the machine learning capabilities, the faster one can create a completely unfair offering in the market, utterly destroying competition. The loop is exponential: it matters a lot what kind of dataset one starts with. The few data heavy companies such as Amazon, Baidu, Google, Mychat and Facebook with few data heavy government agencies such as NSA, tax authorities and health authorities are able to utilize this advantage if they act quickly. However, if one doesn’t have data, one doesn’t stand a chance.
What kind of data is needed? Typically, bigger data than anyone can imagine. But that is not a useful thumb rule. Here is a better one: the best possible data in any industry or market area always wins if there is sufficient machine learning capabilities and a typical platform business model.