AI companies tend to be one of two varieties, either:
- A very science-driven approach, who are typically more academics working on one fundamental problem but serving many industries, say general NLP, or
- Solving problems for a very specific type of customer, and belonging to one specific industry
For brevity, the former can be labelled as horizontal AI startups, or HAS, and the latter vertical AI startups, or VAS.
Traditionally, HAS typically evolved from funded research programs, academia, military, or corporations who perform fundamental science innovation. VAS can be started with less heavy research, thus less funds and rather focused on intelligently applying technology to existing problems. Because of current open source technologies, the opportunity to develop VAS has increased at a rate not seen previously.
Building a VAS
The first rule of starting a VAS is to not focus on building an AI startup, but to focus on solving a problem for a customer or user, applying AI. This may be a slight terminology difference, but the ramifications are drastically different – the former is building technology for its own sake, versus the other is building to solve a customer pain using technology. We see many examples of founders being in love with the technology, when the orientation should be on the customer.
Another reason why starting a VAS approach is important is because by constraining who the customer is, a startup constrains the data in which they need to train their algorithm. By identifying who the customer is, what data is important, and the quality of that data, a startup trains its algorithm in the most efficient manner possible.
One more important reason why a VAS approach is an earlier approach than HAS is because of the competitive advantages it affords. HAS plays into the hands of large corporates like Google, Facebook, Amazon, Baidu and the like, and they have infinitely more human and financial capital to succeed in the long run. VAS, on the other hand, affords a verticalized approach that big corporates likely won’t join in because in their eyes the space is too niche. Assuming that the VAS startup moves quickly, they should build up a very good sustainable advantage and data moat that not produces a sustainable business, but a valuable one as well.
Building a HAS
But can AI startups be purely technology driven, as opposed to customer driven? The answer to that is yes – they can be – but it does require an incredibly strong technical team. Examples of this include Deepmind and Sentient.ai, where the founding and core team members were strong academics or deep AI practitioners with extensive experience. And even then, raising capital for those startups was not an easy path because of the lack of a well-defined path to revenue.
However, if a startup decides to go HAS, there are two conditions to differentiate – one is to focus on the democratisation of the base technology and the other is to create a strong community around consumers of that technology. An example of this is Digital Ocean, which was competing with AWS and Softlayer at the time, but because of its strong community roots and easy user interface, was able to gain traction and build a strong business.
The upside of HAS startups is not to be understated. Every ground-breaking technology comes to a tipping point when it flows out of the hands of its creators and gatekeepers for the overall community. By enabling community involvement and innovation, the base technology becomes more widely used and in turn, more widely adopted. A good example of this is VisiCalc, a spreadsheet application for the Apple II that many argue is one of the prime reasons for the mass-scale adoption of desktop computers. A similar logic can apply to HAS startups.