Why Network Effects Are Important In Robotics Companies

What to look for when you invest.

Happy Sunday and welcome to Technically Sentient! I’m Rob May, a Partner at PJC specializing in robotics and AI investments. This bi-monthly newsletter is a collection of news, research, and commentary on the AI industry.

— Interesting Links —

AI powered virtual influencers are making real money. Bloomberg.

Can Lab-Grown Brains Become Conscious? Nature.

Detecting Covid-19 Through Cellphone Coughs Using AI. MIT News.

Machine Learning The News For MacroEconomics Forecasting. BankUnderground.

Mark Cuban is sponsoring AI bootcamps. WSJ.

Can AI learn common sense from animals? Venturebeat.

Rethinking Attention With Performers. Google AI Blog.

Yann Lecun on why GPT-3 is overhyped. Facebook.

Robotaxis: Where We Are: EETimes.

— Research Links —

Away From Trolley Problems And Towards Risk Management. Link.

Generative Adversarial Networks In Human Emotion Synthesis. Link.

— Commentary —

Building a cutting edge robotics company is difficult. We are at a moment in time where the inputs to a typical robot are changing rapidly. Sensors, arms, cameras, motors, are all getting way way better and way way cheaper at a very rapid pace. This means it suddenly becomes economically feasible to do things with a robot that you couldn’t do before. And it means robotics companies can be built with less capital than a similar company would have required in the past.

But because things are changing quickly, it sometimes presents a problem for robotics companies. Imagine you are building a busboy robot. It can clean off a table in a restaurant in 5 minutes. It’s a bit slow, but human busboys are hard to come by and don’t want to work for the minimum wage most restaurants pay. When humans clear the table though, it takes roughly 45 seconds.

Now say this robotic busboy company has raised $15M from venture capitalists, and can license it under a RaaS (Robots-as-a-Service) model for $250 per month. How should this company think about building a v2 of the robot that is faster and more functional versus selling more of a v1 robot? There are a few factors at play here:

1. A v2 would be faster, and possibly better. Say it could bus in 2.5 minutes, and has the added functionality of being able to load the dishes in the dishwasher.

2. General tech trends around sensors, motors, and other robot parts are advancing at rapid rates in quality, and quickly declining in price.

3. To raise more financing, the company needs to show product/market fit - that people like buying and using the robot.

From a business strategy perspective, this presents a problem. The busboy robot isn’t really defensible as a business model, particularly under a RaaS model. With all the rapid tech advances and declining costs, in 2 years someone can probably build a better robot for a lower price and compete directly with you. To counter that, you have to go build the v2 robot, and push the limits on every technology generation.

But that runs counter to your need to sell what you have and gain market penetration and revenue to continue financing the business. In this case, a first mover advantage is a problem as it validates the market and allows a fast follower to come in behind you and undercut your business. I’ve had this concern with many robotics companies I’ve evaluated for investment.

So how do you get around it? There are a few ways.

1. Engage a modest TAM. In general VCs like to invest in multi-billion dollar TAMs but, if you engage a smaller TAM that is still big enough to build a $100M revenue company, you may discourage competition because of market size.

2. Pre-emptively announce new products. When I worked in the computer chip industry, this was a common tactic. One company would announce "in 18 months we will be releasing a wifi chip for mobile devices with XYZ power consumption and a ABC footprint.” This kind of foreshadowing caused other companies to evaluate if they really wanted to compete for the same market, given the long design cycles for hardware. And it can keep customers loyal if they like your brand and know better things are coming. But, be careful of pushing vaporware. If you are too far off on delivering, it can tarnish your reputation.

3. In some robotics markets, distribution can be an advantage. If you can find and own proprietary channels, that can work. I think this would be particularly beneficial in consumer robotics.

4. Patents may be able to help fend off competition in some cases but, my experience is that when tech fields are changing as fast as robotics is right now, defending patents is difficult. Companies can find ways to accomplish the same thing with other related tech.

5. Move away from a RaaS model to a one-time purchase model. Investors love recurring revenue streams so this may be tough, but, when someone has purchased a robot that still works, instead of licensing one month-to-month, they are less likely to replace it until they absolutely have to.

6. Develop robotic network effects, which I discuss below.

The best way to do this, in my opinion, is to look for robotics opportunities that introduce network effects, so that the purchase decision for the buyer is about more than just the task the robot does directly. I see three ways this can happen.

First, the learnings of one robot can be shared with another robot if they work in the same environment. Imagine if iRobot launched a laundry robot and a dish washing robot to go with the Roomba, and those robots could learn instantly from Roomba how to navigate your house. This kind of technology is a bit off but, will be possible soon.

Secondly, look for use cases where robots operate in small swarms. I’ve seen military combat drones, cleaning drones, and inspection robots for oil and gas tankers as examples of cheap disposable robots that work together in a swarm. Here the real intelligence is in the network of how the swarm behaves together, and that is much more difficult to replicate and the issue I first highlighted - better performance at lower cost because of advances in robotic parts - doesn’t matter.

And finally, look to data network effects from a robots as sensors out in the world. In many agricultural, logistics, and industrial applications I’ve seen where a robot uses machine vision and has a host of other sensors, that accumulated sensor data across locations and environments, and the way it impacts the nuance of robotic performance, becomes more valuable than the hardware of the robot itself. This is another way out of the problem that started this commentary.

The lesson here is, when starting a company in the robotics space, or considering an investment in one, you have to filter out standalone robotic applications that don’t have network effects or some other advantage beyond just being a great robot. Since the market is still early, there aren’t a lot of best practices around robotics businesses, and there are very few executives who know how to build and scale robot companies. Most of the companies are started by roboticists and as a result, many fall in love with parts of the technology or pet problems and don’t think through the larger implications of the business issues they may face.

Thanks for reading.

@robmay