Uber 2.0: Human Self-Driving Cars

There was one line that jumped out of the San Francisco Chronicle’s coverage of the decision of U.S. District Judge Edwin Chen to grant class-action status to drivers seeking tips they claim they are owed by Uber (emphasis mine):1

In deciding whether workers are employees, courts look at how much control the company exerts over them, and how integral the workers are to the company’s business, [Maggie Grover, an Oakland employment attorney] said.

Obviously, you can’t have Uber without drivers,” Grover said. “The court really hit on that.”

That’s certainly the case today, but Uber CEO Travis Kalanick for one has been quite clear that the long-term future for Uber is driverless cars. At the 2014 Code Conference Kalanick had the following exchange with Kara Swisher:2

KS: What did you think of the self-driving car?

TK: Love it. All day long. I mean, look, I’m not going to be manufacturing cars. That’s not what I plan on doing. Somebody’s got to make them. And when those bad boys are made, look, the way to think about it, the magic of self-driving vehicles, is that the reason Uber [is] expensive is because you’re not just paying for the car, you’re paying for the other dude in the car.

KS: Who’s driving.

TK: Who’s driving, right. And so, when there’s no other dude in the car the cost of taking an Uber anywhere becomes cheaper than owning a vehicle. Even if you wanted to go on a road trip it would be cheaper. And so, the magic there is that you basically bring the cost down below the cost of ownership for everybody, and then car ownership goes away. And of course that means safer rides, that means more environmentally friendly, that means a lot of things.

It’s an exciting vision, but even the most optimistic projections for self-driving cars put them several years into the future; today we are still served by what I’m calling Uber 1.0: cars driven by professional drivers. And, this lawsuit notwithstanding, there’s no question that Uber 1.0 is a huge success. I explained last fall why Uber was already on its way to dominating its competition with Lyft in particular, and the company has only pulled further ahead since then.

Still, Uber 1.0 is basically a better, and sometimes cheaper, taxi that is primarily used for taxi-type use cases:3 moving around urban areas, getting home from the bar, going to the airport, etc. To Uber’s credit the service’s coverage and liquidity is often far better than taxis ever were, and the convenience of hailing and paying your driver makes it viable for those who live in urban areas to get rid of their cars completely (as Megan Quinn did).

Trunks and Branches

However, as Kalanick acknowledged, this isn’t a viable option for the vast majority of people. The problem is that there are two types of driving: call them “trunk” and “branch.” The latter happens when there is a critical mass of people and destinations — basically large cities. The former, on the other hand, is about much longer trips: think commuting from a suburb to said big city. It is critical to distinguish these different types of driving behavior for two reasons:

  • The average cost/mile is likely to vary significantly. Specifically, branch (urban) driving is much more expensive due to worse gas mileage, increased wear-and-tear, and especially much higher parking costs. Trunk (commute) driving, on the other hand, likely gets better gas mileage, reduced wear-and-tear, and likely has free parking on at least one end of the trip.
  • The relative cost of identical actions varies significantly due to different trip length. For example, suppose it takes 5 minutes to find parking; if that happens at the end of a 30-minute commute, that means 14% of your total trip time is spent finding parking. On the other hand, if you have to find parking after a 10-minute trip to lunch, that means 33% of your total trip time is spent finding parking.

Given this, you can see why Uber is increasingly a viable alternative to private cars in urban areas: its cost-per-mile is much more competitive even as its convenience advantage increases significantly. But, on the flipside, you can also see why most can’t follow Quinn’s lead: Uber simply doesn’t work for commutes.

Look at the commute data for the average American:4

  • The average American commutes 12.6 miles in 22.8 minutes each way5
  • The average American works 1,789 hours a year, which translates to 224 8-hour days; 224 days X 25.2 miles per day = 5,645 miles/year6
  • Private cars (not including parking costs) cost about $0.50/mile for the average American.7 Uber, meanwhile, prices based on a combination of per-trip flat-fees, a price/mile, and a price/minute; taking a basket of cities8 the cost-per-mile for an average commute is $1.80/mile

Add it all up and commuting with a private car costs $2,823/year, while an Uber costs $10,161. However, this doesn’t include parking: the average American pays $1,300 a year for parking,9 which bumps up the cost of a private car to $4,123/year, which is still a lot less than Uber.


To this point most folks, including those doing these cost analyses, seem to view UberPool as a bit of a novelty at best. Uber promises that the service can save you between 20% and 50% off your fare, which means an annual commute cost of between $5,081 and $8,129; the lower fare is at least in the ballpark of a private car, albeit nearly $1,000 more expensive (and that’s presuming you get the full 50% savings every trip). But seriously, do people really want to go out of their way to pick up another passenger?

Actually, maybe yes, particularly for commutes: I noted above that searching for parking at the end of a long commute requires relatively less time (even if the absolute time is the same); the same principle applies to time spent picking up and dropping off another passenger. An extra minute or two tacked onto a 25 minute commute is much less of a problem relatively speaking than the same minute or two on a short trip. Moreover, the absolute cost savings of UberPool are significantly greater on long trips than they are on short ones. All things being equal I would expect UberPool to be more attractive on “trunk” routes than it is on “branch” routes.

The problem, though, is that all things aren’t equal: UberPool by definition requires lots of riders, which means the service may be doing quite well in San Francisco where, as noted, Uber is already cost competitive with private cars; however, as long as Uber is too expensive for commuting there won’t be enough UberPool potential riders to make UberPool on commutes a reality.

Maybe we’re stuck waiting for self-driving cars.

Uber 2.0

Here’s the thing, though: Uber could have self-driving cars within a year! It just depends on how you define a self-driving car. To a private car owner a self-driving car is a car that, well, drives itself, and as I noted above, the technology is several years off at best.

Uber, though, has a different definition; look again at Kalanick’s comment to Swisher: “The reason Uber [is] expensive is because you’re not just paying for the car, you’re paying for the [driver] in the car.” In other words, from Uber’s perspective, a self-driving car is a car where they don’t have to pay for a driver; the implementation details don’t matter.

With that in mind, think again about the commute problem: right now approximately 75% of Americans drive alone to work. Every one of those solo commuters is a potential UberPool driver, and not just that: because they are making the trip whether they are an UberPool driver or not, they are, from Uber’s perspective, self-driving cars. They are drivers Uber would not need to pay for. This, I believe, will be Uber 2.0: human-powered self-driving cars primarily focused on commutes.

On the surface this is basically car-pooling, but there are some essential differences with this new kind of UberPool:

  • Uber has solved (or is solving) the coordination problem that plagues any highly dispersed set of actors; instead of finding a carpool rider and arranging pick-ups simply open the app
  • Uber has a ready-made network of potential drivers and riders who already use and, more importantly, already trust Uber. Moreover, the (human) self-driving UberPool network doesn’t need a huge number of riders to ensure liquidity: after all, all of those (human) self-driving cars are making the commute regardless
  • Uber 1.0 — the urban branch network — solves a major problem with carpooling: the fact that the rider doesn’t have a car at their destination. But as long as their destination is an urban area, that’s ok — they don’t want a private car anyways!

What is even more impressive, though, are the economics of this potential service:

  • 80% of the cost of an Uber 1.0 ride goes to the driver. Take that away and Uber’s $1.80/mile cost plummets to $0.36/mile
  • Uber needs to then give some of those savings back to the UberPool self-driver, but significantly less than it needs to pay a professional driver. After all, the UberPool self-driver is making the drive anyways: whatever payments they receive from carpool passengers is effectively found money

Let’s presume Uber pays UberPool self-drivers $0.36/mile (basically, Uber takes half of the total fare): that means for UberPool passengers the cost of a trip is $0.72/mile, or $4,064/year for their commute. That is less than the cost of a private car once you include parking!

Moreover, Uber would likely be a lot more aggressive in pricing; after all, people who start using (human) self-driving UberPool cars are also more likely to use Uber 1.0 in the short-run, and completely give up private cars in the long run. Ultimately, the more that Uber can make (human) self-driving UberPool cars cheaper than private cars the more the company stands to gain in the long run.

Whither Google?

Back in February there were a flurry of stories about Google potentially competing with Uber, with the presumption that their self-driving cars would be the trump card. As I wrote at the time:

Google could build their own Uber. True, they would start way behind when it comes to consumer liquidity, but if their cars are all driverless, that doesn’t matter. The entire reason consumer liquidity is important is because it attracts drivers to the service, but a driverless car simply does what it’s told. And, given that >70% of the cost of an Uber goes to the driver, Google’s service could have significantly lower prices right off the bat, and, as I noted in Why Uber Fights, truly lower prices are one of the ways that Uber’s network advantages could be overcome.

Here’s the thing, though: a (human) self-driving UberPool car is cheaper than a Google self-driving car. Remember, all of the costs associated with a (human) self-driving UberPool car are sunk: the car and driver are making their commute whether or not they have a passenger, which means all of the various costs should not be ascribed to Uber (in contrast to a professional driver who should discount all of their depreciation, gas, and maintenance costs from their earnings). Basically, (human) self-driving UberPool cars have zero cost from Uber’s perspective (obviously, as noted above, Uber needs to provide an incentive to the driver to pick someone up, but it’s very reasonable to presume that could lower on a per-mile basis than Google’s costs).

Google’s self-driving cars, on the other hand, would have all of the costs associated with professionally-driven cars: depreciation, gas, and maintenance. After all, those are costs that are accrued for the sole purpose of providing a transportation service. And while it’s true that Google could afford to eat those costs, they would still have to build cars at scale and overcome Uber’s built-in network.

To be sure, self-driving cars would be a superior alternative to Uber 1.0’s professional drivers providing branch service; were Google to compete they could obviously start there. Ultimately, though, I expect them to back off: a technological advantage, which Google likely has, is by definition temporary; a network advantage, though, only grows with time.

Public Transport?

Last week BuzzFeed reported that Uber was experimenting with “smart routes”:

The idea is drivers will be able to both pick up and drop off passengers along a specific route, which in turn allows them to quickly pick up their next passenger…Uber Smart Routes are similar to traditional bus routes in that they follow a predetermined path between two points, but they differ by allowing passengers to request rides on demand. The catch, of course, is that customers get themselves to a Smart Route before they can use it. So it’s not exactly door-to-door service. For this reason Uber is discounting Smart Route rides by $1 or more.

First off, this concept is clearly applicable to the (human) self-driving UberPool network: one could imagine a 101 smart route down Silicon Valley, or another on El Camino Real. Many people, though, objected to the fact that, well, smart routes look like bus routes, and wouldn’t it be better if we spent more money on public transportation?

The answer, in my opinion, is no. I am a long-time supporter of public transportation, and feel blessed to live in a city with an extensive, efficient, and reliable subway and bus system. I wish that the U.S. in particular had invested more over the years in its infrastructure. However, if everything in this article comes to pass — which in the long run, will mean scores of people giving up private cars — many of the benefits of public transportation will be a reality: less pollution, less need for parking, fewer drunk drivers. Moreover, there will be some real advantages to a car-based network: Uber will offer far better service to less dense suburbs and exurbs than would ever be feasible for public transportation.

That leaves the less fortunate, which in itself is telling. Joseph Stromberg wrote a compelling piece on Vox that argued that the entire reason U.S. public transportation is so deficient is because the U.S. has treated public transportation investment as basically a welfare program, resulting in too-low fares that make efficient service impossible. Stromberg concludes:

Nowadays, many local politicians don’t see transit as a vital transportation function — instead, they think of it as a government aid program to help poor people who lack cars.

In the future I’m describing, though, no one owns cars! Wouldn’t it be much simpler and more efficient to simply subsidize the needy directly so they could use the same transportation network as everyone else?

That, of course, gets to the ultimate Uber end-game: I more than ever believe they have the potential to completely transform transportation in the United States if not the world, and it’s difficult to see them doing so without extensive regulation. One suspects Uber’s courtroom drama is only beginning.

  1. Class-action status was not granted with regard to the question of expenses, although the plaintiffs can resubmit their request 

  2. If you’ve never watched the video, the exchange immediately after this about Google is genuinely funny. It’s at the 19:20 mark 

  3. I absolutely do not believe that Uber succeeds simply by being cheaper; the experience is significantly better 

  4. This post that attempted to model when it would be viable to go Uber-only was particularly helpful, especially when it came to linking to original sources. However, that article did double-count the commute distance, throwing off its conclusions 

  5. Source: U.S. Department of Transportation Federal Highway Administration (FHWA)  

  6. Source: OECD  

  7. Source: the American Automobile Association and FHWA 

  8. Specifically, Los Angeles, San Francisco, Houston, Chicago, and New York City 

  9. Although obviously significantly more in large cities: $4,500 in San Francisco, for example, and $2,520 in Los Angeles