Market timing and some thoughts on self-driving car

I published a short version of this post focusing on self-driving car here

Before embarking on any project, we often have to convince ourselves and others that the project is heading in the right direction. The first thing to consider is whether there are competitors. If there are already giants in the field, then the project is likely a nonstarter. You are probably unlikely to start a food-delivery startup now, but what about projects like a self-driving car if you have right expertise?

Winner-takes-all in technology

In technology-driven businesses, it is often winner-takes-all because the technological advantage helps break into a market to access to more customers and more resources. This allows the winning company to improve its technology/product better at a faster rate than lagging competitors, which brings on more advantage, resulting in a positive feedback cycle. Usually, in many enterprises, such a virtuous cycle has to stop at some point, some earlier, some later. If the business in question is a local family-owned restaurant, the virtuous cycle has to stop when the restaurant is full every day. If the business is a search engine, the cycle can go on for a very long time. To the extent that it develops an AI to call up restaurants to check their opening hours. Just so that users can now search for information not on the Internet

So am I late?

If it’s winner-takes-all, then it is really bad to be late. Hence, the question: “am I late to X market?”, In the simplest case, if your technology is 10x better than anything out there in a very visible way, in a fast-growing market that already exists, then you are not late. An example of such dominating technological advantage is Google’s PageRank. Internet was growing very quickly then and the 10x improvement of PageRank was evident to users. PageRank essentially erase any head start that the other competitors may have.

However, it’s often not so clear at the beginning. With the benefit of hindsight, it is easy to point out the 10x advantage in successful projects. But right now, you might have something interesting but not sure where it fits in. In that case, it’s not so helpful to think about whether you are late to a market. The question a priori assumes that the market is known and the product is well-defined. It’s not always clear whether the product and market assumptions hold. Even if the market leader has found a product-market fit, it is more helpful to think of it as one of the ways things could work, probably not the only way. In fact, worrying too much about the competition can have a devastating effect on a project by pushing you towards your competitors’ game, where they have all the advantage. Projects might end up pursuing minor differentiating factors on paper that don’t matter. If Google were to be worried about Yahoo back then, it might have built a bigger Internet directory instead of the search engine that we know. One strategy towards competition is to understand the assumptions or shortcuts made by others in the business and attack the least valid one. Better still, envision your own version of future and change the way things are done.

Evolution of self-driving car development

The approach people takes toward the self-driving car and their assumptions they made has evolved throughout the years with the maturity of the technology. At the very beginning, teams at DAPRA urban challenge threw tons of expensive sensors for perception and hard-coded just enough control logic to get through the driving course. The assumptions were that the expensive sensors setup could be simplified, the control algorithm could be refined, tricky scenarios could be tackled later. The key question was whether driving can be highly automated.

In 2018, the question about self-driving is no longer about whether it can be highly automated. While L5 self-driving (able to drop the car anywhere on earth and it will drive itself right away) might still be decades away, current level of driving automation can be good enough to be useful in many situations. There are still many issues with today’s technology such as costly sensors, safety, and the need to map (and maintain updated map) of the self-driving routes. In general, in an area or route where the traffic is light and the roads are wide and clear, given enough mapping, fully automated driving along predefined routes should work pretty well. The question today is then how to deploy current driving automation in a scalable manner.

With the current state of self-driving technology, for each new location, the self-driving car company has to build the map, make sure location-specific traffic requirements are handled, and test out the routes before rolling out to the public. I think it is useful to draw parallels with how ride-hailing companies expand given that their service is also location dependent. The dynamics of ride-hailing expansion could give hints on how widespread deployment of self-driving car may happen. In ride-hailing, the usefulness of the platform depends on the density of drivers and riders in an area. The higher the density of drivers and riders, the easier a rider can get a ride, and a driver can get a job. The platform that attracts more users wins and the losing platform become less and less useful. Hence, a ride-hailing company typically invest heavily in customer acquisition to get both riders and drivers onto its platform and to wrestle them from competing platforms. While ride-hailing is best known in the media for its “disruptive” technology, the main battlefield for these companies is still capital-intensive customer acquisition, not about technology

But for self-driving cars, technology is their competitive edge more than ever. Let’s consider how self-driving companies may compete in a new location. Self-driving car companies first have to incur a fixed cost in terms of mapping and testing when deploying to a new location. Unlike ride-hailing apps, they can directly control the number of self-driving cars on the road, which solves the initial problem of establishing a two-sided market. What they cannot directly control is the serviceable routes in an area, as some roads may be just too hard to navigate automatically. The more pickup and drop-off points the self-driving cars can serve, the more useful is the self-driving car network. Technology affects all aspect of self-driving car expansion and competition. Better mapping and adaptation technology can drastically lower the fixed cost and time it takes to expand to a new location. Better perception technology lowers the fixed cost per vehicle as they can use cheaper sensors. Better self-driving technology enables one to navigate in difficult places where one cannot go otherwise, widening the reachable points in the network, and hence increasing the usefulness of the network. Hence, unlike ride-hailing apps, the competition between self-driving car companies will not be about who has a larger war chest.

If such a dynamics in widespread self-driving car deployment is true, it is more imperative than ever to invest in the right scalable technology. This makes it hard to justify building a generic self-driving car using whatever available technology today regardless of cost and scalability. It is too late to approach the problem from a 2008 perspective in 2018. Building a self-driving car prototype today might still attract attention in communities that have not seen one. It may attract business partnerships in industry adjacent to self-driving such as a smaller automated logistic robot (or automated guided vehicle), which can be valuable in its own right. But having a generic self-driving car is not enough to break into the next step of the self-driving game.

Nothing wrong with being opportunistic, but…

The question “am I late to X market” implies an opportunistic view of the market where one seeks to make quick bucks by jumping onto the bandwagon. There is nothing inherently wrong with being opportunistic. But it is how you evaluate opportunities that count. In fast-growing domains, what is valuable changes rapidly due to new information or competition. The crowd usually converge on an outdated view of what is valuable. The mistake usually associated with being opportunistic is to chase what looked like valuable in the past instead of what is going to valuable. An entrepreneur has to be more forward-looking than that. Indeed, it is sometimes possible to win by following the path of a successful competitor. After all, they have done the hard work blazing a trail and it seems less risky to follow that trail than taking a new path on your own. One might say that ideas are cheap, execution is everything, and hence one can out-execute the competition. But an important part of execution is to deeply understand the market and maximize your advantage in delivering something useful.

The key to breaking into the self-driving car market today is to focus on improving the technology for scalable commercialization. This might mean working with cheaper sensors but in lower speed or more structured environment. It can be better mapping and testing process or hybrid sensors. This might mean experimenting with smaller transportation modalities than a full-size car. It may turn out the smaller modalities is the right answer to automated personal transportation. Or the project can be valuable if it can generate technology that lowers self-driving car deployment cost.

While timing the market is hard and it’s not useful to over-worry about, it is also arrogant to think that one person or company can single-handedly change the world as we know it regardless of everyone else. Almost all out-sized success story in technology is a result of culmination of small innovations, tons of hard work internally and larger changes in the external environment. They ride on waves of technological changes. Technology innovators should avoid short-termism and sloppy thinking, and should look further into the horizon of the technology changes. The Amara’s law put it best “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”