My take on Ben Thompson Aggregation Theory.
When the marginal cost of serving a new user is low companies tend to grow big. It makes sense to split the fixed cost between a bigger number of customers when the additional cost of serving new customers does not matter. The easiest way to grow big is to shed off all parts of the business that does not scale, outsource them outside of the company. [I m looking for a good quote here — but it feels like common sense, microeconomy 101]
The path to low marginal costs is automation. This has not changed since the steam engine, but now we have more things that can be automated and also automation of information processing is much more extreme than automation of physical work. Among other things automation lets companies have a direct relationship with huge number of customers and internet provides zero cost automated distribution. But in information technology we can also have network effects — where serving new customer improves the utility for other customers — this is as if the marginal costs were negative. This leads to a new kind of business which outsources all of the physical and other hard to automate processes and provides by itself only customer interface. It becomes an aggregator.
If the outsourced part is fully standardized you can have a digital marketplace business like Uber or AirBnB. This is good place to be — you have commoditized your complement. It is even better when the non-automated part of the service is provided by the users themselves — like content in social networks or web search. These are called Supper Aggregators (or Level 3 Aggregators) by Ben Thompson — everything automated. In these cases you need an additional part of the business to be the profit centre — advertising usually. Fortunately this also can be automated with self-service for the advertisers.
Digital marketplaces are usually Level 2 Aggregators — because they cannot automate “bringing suppliers onto their platform”. For example Uber needs to do background checks and vehicle verification for each new driver. But this is not universal rule, they can also be Supper Aggregators.
There are also Level 1 Aggregators — where the non-automated product/service part is not standardized enough for a marketplace, but it still can be outsourced. This is the case of Netflix buying content, and then serving it in a fully automated self-service.
There are additional complications — especially in the interplay between local and global.
AirBnB is a better aggregator than Uber — because customers in hospitality market come from all over the world and need something that will be trusted (and recognized) globally — while taxi users are mostly from local population. A local taxi company can easily compete with a global one, but short term accommodation needs a global marketplace.
Drivers (for Uber and Lyft) and apartment/room owners (for AirBnB and Booking.com, Expedia and others) are not bound to any marketplace and can choose between offers. This is a problem for the aggregators because it lets their providers commoditize them. In hospitality market there are Channel Managers which are kind of aggregators of aggregators — they aggregate the marketplaces as marketing channels for hospitality providers. It would be very useful to analyse when that can happen and especially when you can automate it like the Channel Manager software does.
But not every startup is an aggregator — sometime they can automate such a big part of a job that they have no need for outsourcing. For example Codility (disclosure — I am an investor there), found a way to automate evaluating programmers and does not outsource any part of its main job. Codility has also went a long way automating content creation (programming questions in their case), but we are still to find out whether in this case the ultimate solution will rely on outsourcing or further deep-tech automation.
Related readings: https://stripe.com/atlas/guides/andrew-chen-marketplaces