In my post on OnDeck (ONDK) I included the collection of user data as key competitive advantage.
Specifically, I thought ONDK benefited from “data network effects.” What are data network effects? As an online platform (such as ONDK< Google, Facebook, Amazon, Pinterest or Twitter) achieves scale and gains users, it acquires more data. This data leads to product improvement, which leads to more users and, subsequently, more data. The process repeats. According to some, this dynamic leads to an unbreakable positive feedback loop that makes effective competition impossible. In fact, I included this chart illustrating the data network effect feedback loop:
I bought this line of thinking without much scrutiny. Since then, I have reconsidered the basic premise. At issue is whether the collection of large amounts of data—sometimes referred to as “big data”—and in particular, data collected from users, can lead to markets “tipping” to dominant online platforms. In other words, does the accumulation of data by Internet companies create a moat because new entrants will not be able to compete effectively with the first mover in the marketplace?
The accumulation of data does not seem to create a moat for the following reasons:
- Data is Non-Exclusive
The basic small business behavior and performance data upon which ONDK relies is not a finite, excludable resource like a rare commodity. No firm can control or exclude others from using it. Indeed, ONDK does not claim to rely solely on proprietary data. It states that it uses 100 external data sources. Presumably these data sources are available to competitors as well.
Online services also suffer from the multi-homing problem. The multi-homing problem occurs when a user uses more than one online platform. Due to the lack of switching costs, I can easily use Bing and Google if so choose. As a result, both search engines can collect data on me. Similarly, potential borrowers can access ONDK, Kabbage or any of a number of other online lenders. ONDK’s loyalty program will create some switching costs, but the market is so huge that even in the best case it will restrain a small minority of users.
- Data is Non-Rivalrous
A rival good is a good whose consumption by one party prevents simultaneous consumption by another party. Data is non-rivalrous. The collection and use by one party does not detract from collection by another party.
- The Marginal Returns on Data Diminish Rapidly
Scale benefits are subject to diminishing returns. Accuracy generally increases as a square root of the sample size, so doubling the sample size equates to roughly a 41% increase in accuracy. Hence the rapidly declining returns to scale.
- Ideas Matter More Than Data
The data networks feedback loop is not an iron clad rule of nature. Competition is not based solely on how much user data a company can collect. If it was, the Too Big To Fail Banks would already dominate online small business lending.
In fact, upsets occur all the time in online competition. Google replaced Yahoo as the leading search engine despite Yahoo having years of user data and “first mover advantage.” Facebook overcame MySpace despite the latter company’s data lead. There is no reason to believe a new entrant in the small business lending space could not invent a better mousetrap and surpass ONDK.
- The Value of Data Decreases Rapidly Over Time
The value of data companies collect decreases rapidly over time So the data ONDK collects now, maybe worthless in 3 years. If the value of data didn’t decrease rapidly over time, back testing of stock performance would yield a magic formula for success. That would make all of our lives a lot easier. Unfortunately, that magic formula does not seem to exist.
- Are Data Network Effects Really Supply Side Economies of Scale?
Network effects are also known as demand side economies of scale. That is, as more users use the network, the quality of the service increases and, therefore, the demand for the service also increases.
Supply side economies of scale means that the per unit cost of production drops as the number of units produced increases.
Here, I made the classic mistake of confusing demand and supply side economies of scale. Borrowers don’t care if more borrowers are using it. It’s not like a social media network like Facebook where it is critically important how many people are using the network. Instead, as ONDK gets bigger, the quality of the service improves – a supply side scale effect.
Correcting this mistake than leads to the next question, what is the minimum efficient scale to compete in the small business lending space? How much data is needed to hit critical mass? Is it 10 years of data? 5 years? 1 year? Consider this slide from Hal Varian, chief economist at Google:
The graph shows the correlation between the ability of the algorithm to solve problems and the amount of data collected. Once the threshold amount of data is collected, the quality of the algorithm plateaus.
Where is ONDK on this curve? Where is Kabbage? Where are other entrants? It’s impossible to know.
- Is Growth Profitable?
It goes without saying that more customers and more revenue is better than the opposite. But the question is whether the growth is profitable. Does the cost of producing incremental quality decrease with scale?
Here, more customers can give you better data which can give you better understanding of credit risks, but it does not lead inexorably to an increase in profits. An increase in customers has other impacts, including an increase in costs. ONDK must spend money on advertising and marketing to acquire customers. It needs personnel to process loan applications. It needs systems to collect data, databases to manage data, analysts to understand data, experiments to validate this analysis and management to implement changes, etc. It bears pointing out that ONDK has never had a year where it made money on a GAAP net income basis.
The positive post on ONDK was a mistake. Mea culpa. “Big data” is a buzzword and has lost connection to reality – much like “dot-com” for those old enough to remember the 1997-2000 bubble or “conglomerate” in an earlier period. A company does not become more valuable simply because it has access to data.
The core problems were trusting an unproven and untested business model and not challenging my basic thesis enough. It will not happen again. Luckily, my readers largely chose to ignore this post. The ONDK got far fewer views than my normal posts.