The questionable economics of autonomous taxi fleets

Alphaville has long held the suspicion that, despite the hype, the economics of a self-driving fleet of taxis, as an alternative to owning a car, simply won’t work.

A new paper out Monday, written by researchers at the Massachusetts Institute of Technology and exclusively shared with FT Alphaville, agrees. It suggests that, at current prices, an automated hive of driverless taxis will actually be more expensive for a consumer to use than the old-world way of owning four wheels.

Drawing on a wealth of publicly available data, Ashley Nunes and his colleague Kristen Hernandez suggest that the price for taking an autonomous taxi will be between $1.58 to $6.01 on a per-mile basis, versus the $0.72 cost of owning a car. Using San Francisco’s taxi market as its test area, the academics examined a vast array of costs such as licensing, maintenance, fuel and insurance for their calculations.

The news comes as the arms race to deliver an autonomous taxi service reaches full-speed in Silicon Valley. Both Lyft and Uber have committed hundreds of millions of dollars to developing the technology, with Uber chief executive Dara Khosrowshahi declaring at a recent FT conference that “if there’s one big goal for Uber it’s to replace car ownership”. Along the same lines, Alphabet-owned Waymo has been operating a robo-taxi service in Phoenix, Arizona since December.

A robo-taxi service have two main economic flaws, according to Nunes and Hernandez.

First is what the two academics refer to as “capacity utilisation” — the amount of time an autonomous vehicle is carrying a customer. According to the paper, the taxi occupancy rate stands at 52 per cent in San Francisco. Whereas in car ownership fuel and usage are directly correlated, a taxi is only being used around half the time.

In San Francisco, traditional taxi fleet operators circumvent this issue by charging drivers a set fee, which covers their administrative costs, in return for leasing one of their cars. As Nunes and Hernandez put it, this means “existing taxi business models ensure revenue generation independent of capacity utilisation”.

However an operator of a self-driving fleet is in effect the driver, and therefore must either absorb the cost of when the vehicle is not being utilised, or pass it on to the consumer in the form of higher prices. At a 52 per cent occupancy rate this is particularly problematic, in part because the operating costs of ensuring the safe running of a self-driving fleet will be high, mainly due to the human oversight required. A driverless taxi, it turns out, will not be people-less.

After inputting costs such as wages alongside overheads such as office space and equipment Nunes and Hernandez suggest it could cost an operator anywhere between $2.35 to $0.05 per mile depending on the number of cars being monitored by one supervisor. (The duo put the limit of cars that can be watched by one supervisor at 50, which given the safety implications, seems high.)

Robo-taxi believers might counter that with improvements in technology. For instance, an AI programme could monitor thousands of cars, spreading the cost burden. But as the cost range above suggests, the marginal cost savings from an operator monitoring multiple vehicles diminishes significantly past a critical point, which Nunes and Hernandez pin at around 10 cars.

Here’s a chart from the study showing just that:

A fleet operator could also move the needle by increasing the utilisation rate. 55 per cent, after all, is pretty low. And with recent developments in location technologies, matching supply and demand will surely be more efficient.

However, Nunes and Hernandez suggest that even with a 100 per cent utilisation rate, the cost of a self-driving taxi will still be higher than the cost of owning a car.

The other option for robo-taxi operators to achieve parity with the cost of vehicle ownership is, of course, to accept lower profits than what is being currently made in the marketplace. In their research, Nunes and Hernandez found that for each mile driven, San Francisco fleet operators currently make around $0.27 of profits on a $3.35 fare. A rather paltry 8 per cent margin.

Uber, Lyft and the other hopefuls are no stranger to losses, so this is a possibility. We called up Nunes to ask if he thought this was reasonable:

An analyst at UBS actually posed that question to me a few weeks ago. Of course fleet operators could accept lower margins. But then why would a venture capitalist, or a company, invest in a self-driving fleet when you can just run a traditional, human one with higher margins?

It's a good point. But then again, investors have been quite happy throwing money at lossmaking businesses in this cycle. So there's no guarantee they'll stop now.

We also noted a crucial assumption in Nunes and Hernandez’s calculations that’s worth highlighting: all of the costs are based on a journey being taken by only one individual.

But with the pooling of rides common place among the ride-sharing platforms, surely the costs come down significantly if multiple passengers are splitting a ride? Here's what Nunes had to say on the phone about this caveat, which they also highlight in the paper:

According to the data, current car occupancy in the US is 1.67 people. So you’d need a significant increase in riders, around 30 per cent, to move the cost per ride close to the current level for car ownership. More passengers means a potentially longer journey, making the service less convenient, and the data shows consumers have an aversion to sharing for-hire occupancy travel, partly thanks to this time uncertainty, and partly down to privacy.

Opportunity costs, such as the value of someone's time when they're driving, is worth considering here, as self-driving cars will theoretically free up time to be spent on other, more productive pursuits. (Such as playing Candy Crush and avoiding eye contact with strangers.)

However, perhaps in anticipation of this critique, the two researchers have come up with a per-mile value for this lost time of $0.89, citing the Department of Transportation's own guidance on the issue.

But lost time is not the only opportunity cost of cars. There are carbon emissions, which are contributing to climate change and damaging our lungs, and then there are automobile crashes. According to Nunes and Hernandez's research, these accidents account for 40 per cent of all “unintentional injury" deaths in the US. It is fair to argue, with a few caveats, that self-driving cars could make a meaningful dent to both of these negative externalities.

Yet after reading Nunes and Hernandez's research, perhaps investors in the space will be asking themselves if these businesses will capture any of this residual value. More likely however, is that this will be yet another instance of that amazing trick where capitalism, gripped by techno-optimism and euphoria, ensures everyone wins. Except the shareholders.

Related Links:
Autonomous Vehicles and Public Health: High Cost or High Opportunity Cost? — Ashley Nunes and Kristen HernandezDo the economics of self-driving taxis actually make sense? — FT Alphaville
SD cars and productivity — FT Alphaville
Uber's conflicting self-driving fleet vision — FT Alphaville

Copyright The Financial Times Limited 2019. All rights reserved. You may share using our article tools. Please don't cut articles from FT.com and redistribute by email or post to the web.

Read next:

Read next:

Lookout, there’s a dollar crunch!

FT Alpha Tweets