We have been thinking about how to evolve our investment strategy recently. I will write about the full process when we’re done and I’ve got a better sense of which bits have worked and which haven’t, but for now I want to highlight a post by another VC which highlights a lot of the methods we like to use when thinking about the attractiveness of potential focus areas.
The post was written by Bradford Cross, partner at Data Collective. Superficially it’s a listicle with Five AI Startup Predictions for 2017, but you don’t have to read very long before finding some good structured analysis and original thinking.
It turns out that four or Bradford’s five predictions are about things that won’t work and one about something that will work. Each of his points has generalisable lessons that can be applied to analysis of any potential investment sector.
- Bots go bust – main reasons: bot interactions are utilitarian and don’t meet our emotional needs, and for most use cases they are less efficient than other UI paradigms (e.g. apps and menus – note that Facebook has just added menu features to Messenger).
- Deep learning goes commodity – main reason: the number of grad students with deep learning skills has mushroomed and the premium paid for deep learning acqui-hires will fall because companies now other options for bringing in talent.
- AI is cleantech 2.0 for VCs – main reason: cleantech failed as an investment category because it’s a cross-cutting societal concern with a self important save-the-world mentality and not a market. AI has similarities, albeit the self-important element is about forming ethics committees and saving the world from the fruits of it’s own labour – super intelligences that destroy humanity and robots that take all our jobs.
- Machine-learning as-a-service dies a death – main reason: machine learning APIs are two dumb for AI experts and too difficult for AI novices. They don’t have a market.
- Full stack vertical AI startups actually work – main reason: low level task based AI gets commoditised quickly whereas vertical AI plays solve full-stack industry problems with subject matter expertise and unique data which make them defensible.
The generalisable lessons here are:
- Use cases are paramount to good investing (ref points 1, 3 and 4). Bots are failing because they don’t solve any new use cases and are worse at their job than other options. Horizontally focused investment themes are tough because they don’t start with use cases. Machine learning APIs aren’t solving a problem for anyone. Good candidates for investment focus areas have easy to understand use cases – e.g. I buy from ecommerce companies because it’s more convenient and the range is better.
- Valuable businesses have strong barriers to entry (ref points 2, 3, 4 and 5). Deep learning, and AI more generally, got hot in part because talent was scarce. This reached the point where $m per PhD was talked about as an acquisition metric. However, talent is not a barrier to entry over the long term and neither is clever implementation of new algorithms. Proprietary data and uniquely trained models on the other hand, can provide a basis for high margins over the long term.
- Hype is dangerous (points 1, 2, and 3). Hyped sectors draw in lots of VC dollars which drive valuations up, creating an illusion of success which brings in more VC dollars (sometimes spurred on by M&A). It is possible to make quick money from investing in startups in hyped markets but it’s a lottery. Moreover, all the mania often causes founders and investors to lose their focus on use cases. Unsexy is harder work, but it wins in the end.
- Good focus areas allow for shared learning (point 3). One of the reasons that cleantech was a difficult place to make money is that there was little in common between different cleantech companies. Solar, wind, and biofuels, for example, all have very different technologies, different customers and different company building best practices. Mobile games, in contrast, has been a successful investment focus for many investors because key disciplines around game mechanics, monetisation and marketing are common across companies.
Many VCs are opportunity driven. Their primary strategy is to work on building their networks and then they invest in the best of what they see. Our belief is that focusing yields better results because deep understanding of a sector leads to better decision making and a greater ability to help entrepreneurs succeed. However, focusing is hard. It takes deep thought and hard work to find interesting areas and then it takes strong discipline to stick to your strategy. Focusing is also risky. If you choose a bad area to focus on at a minimum you will look stupid and if you don’t course correct in time you will have a bad fund. Still, if venture has taught me anything it’s that fortune favours the brave