Wednesday, August 6, 2014

What taxis' operating costs can tell us about driverless cars' potential

I already wrote of my skepticism on the driverless revolution. However, one thing annoyed me in the earlier post I wrote: my estimate on the cost of driverless taxis was really a rough estimate. So I did a bit more research and found something that I think is a good way to estimate realistic prices  The easiest comparison is to current taxis, because what is an automatic car for hire but a taxi without a driver?

So the current costs of taxis are the best way to estimate what automatic taxis would cost. Too often, estimates of costs are grossly unrealistic, essentially presuming there is no overhead cost, no dead heading, etc... If we estimated bus costs the same way, I think we'd estimate them to be peanuts too.

Anyway, here are some sources I've found:

In Québec, the average independent taxi owner pays 10 000$ per year in fixed costs for his car and 100$ per day to run it. Supposing an average of 260 days of work, that's a total of 36 000$.

Meanwhile, the average mileage of a taxi in Québec is 67 730 km (42 000 miles)

So that gives us a total of 0,53$ per km (0,85$ per mile). This, as far as I can tell, isn't just the mileage with passengers inside, but the total mileage. With dead heading, the cost would be much higher. If we suppose 30% dead miles, which seems to me the realistic minimum we can expect, the real cost is 0,76 $ per km, 1,21$ per mile. Not that far off from my initial rough estimate of 1,30$ per mile.

That is the cost to run a taxi, not including the cost of the driver himself and his income.  It tends to include taxi licenses, which are somewhat expensive. We can debate whether or not driverless taxis will require licenses or not still, I think they would, if only to control the supply of them and to avoid them overrunning the streets. A taxi license is necessary to benefit from the right to use some taxi only parking and staging areas. 

I don't think driverless cars will reduce these costs. My estimate of driverless car fares seems accurate.

An aside, we tend to think the more used a car is, the less important fixed costs are per mile... which is somewhat true for regular commuter mileage, 15 000 to 30 000 kilometers a year (10 000 to 20 000 miles) but truth is, when you use a car too much, you just end up scrapping it earlier. The wear and tear of the excessive use of it reduces its lifespan and imposes more maintenance and repairs. For instance, a 3-year-old car with 60 000 kilometers on the counter (35 000 miles) and another with 50 000 kilometers (30 000 miles) would have roughly the same depreciation and maintenance costs, but another 3-year-old car with 210 000 kilometers on the counter (130 000 miles) would be worth much less and you'd reasonably expect maintenance and repair costs to be much higher. So there is a limit to how cheap fixed costs can become as a car is used more and more. Depreciation is a function of time and mileage, at regular mileages, time dominates, but if the vehicle is used excessively, mileage becomes the dominant factor, eliminating savings from greater use of vehicles.

I also found interesting breakdowns of costs for taxis in Canada and Australia

First, here is a document from Ottawa about taxi cost profiles, where the fares go.
Ottawa taxi cost profile
Now, taxi fares tend to be composed of three different factors: a base fare, a waiting fare and a distance fare, so bringing it down only to mile is bound to be a bit difficult with some suppositions along the way. Rather than do that, I'd go by my personal experience, my one recent taxi ride was 2,33$ per km (3,73$ per mile). If the driver's earnings represent 51% of that, that means the net cost of the taxi was 1,19$ per km (1,90$ per mile). If we choose to exclude licensing costs too, the cost would have been 0,92$/km (1,47$/mile). This is actually over my original estimate.

I also found a document from New South Wales in Australia which has a more detailed index:

Cost breakdown of taxi operation in New South Wales
Now, you can pick and choose what you keep and what you remove. The most pessimistic approach would be to just remove the wages, around 42% (averaging county and city), that would leave costs at 1,37$/km (2,19$/mile). The most optimistic approach would keep only the fuel, the other drivers' costs, the operator's salary equivalent, the maintenance costs, the vehicle lease payments, the network fees and perhaps half the insurance (supposing insurance reduction due to automation... still, with the likelihood of vandalism in unoccupied vehicles, will insurance really be down?). That would leave 35,5% of current costs, or 0,83$/km (1,32$/mile). Almost exactly my initial estimate.

I found another document on Canadian taxi costs, but they aren't sensibly different from the Ottawa cost profile.

This was just essentially an annex to my post on driverless cars. I have not been challenged, to my knowledge, on my initial cost estimate of driverless taxis, but I've been challenging myself on this, wondering if I got it right. I think this approach of looking at current taxi operating costs is a much more sound way to approach the issue rather than compare costs to a private vehicle's costs and try to build up an estimate from the ground up without considering the costs to upkeep the entire taxi system in the meanwhile.

I still think the economics of driverless taxis make them unlikely to replace transit or private vehicles on a mass scale, though they would make taxis much more affordable, yet still much more expensive than alternatives, and that taxis as a whole don't scale well.


  1. Here in Mexico City, I usually pay around MX$12 per km. For a ride to the airport at 3am I recently paid MX$150 for 10 km in a secure on-call taxi.

    That's about US$0.90 to US$1.15 per km. The market is extremely competitive and known for being the cheapest big city taxi market in the first world. I doubt anyone is going to be running self-driving taxis any cheaper than that.

    1. Thanks for that input. I agree that it would be hard to offer significantly cheaper fares. From what I can find, Mexico City taxi drivers have pretty low wages already. I've seen estimates of 100 to 200 pesos per day (10 to 20 USD).

    2. This comment has been removed by the author.

    3. Mexico City taxi drivers are not notably wealthy people, and I know only one personally, but I can say that MX$100-200 per day is well below the typical net income. It wouldn't surprise me if that is what SAP sees on their income tax forms, though.

      An apartment in a middle class neighborhood, a couple kids put through college, and even an occasional vacation abroad are within the reach of hard working taxi drivers whose spouses bring in a little extra income, according to consumption patterns in my social network.

      I just don't think that a self driving car is going to operate cheaper than a Mexico City taxi overall. The cars are small sedans, the maintenance is done by personal contacts or drivers themselves better and cheaper than any fleet operation, the regulations and taxes are minimal, there's no expensive sensor radar and lasers to maintain, and wages are cheap without being poverty level.

      And still the taxi costs 10x what the subway, light rail, BRT, dollar vans, and commuter rail cost.

  2. "I still think the economics of driverless taxis make them unlikely to replace transit or private vehicles on a mass scale, though they would make taxis much more affordable, yet still much more expensive than alternatives, and that taxis as a whole don't scale well."

    Yeah, this seems exactly right. I'm not even sure driverless taxis could be cheap enough at any cost to widely replace privately owned motor vehicles, and that does seem to be borne out in Mexico City, Mumbai, Hanoi, etc., as I wrote about. In spite of extremely low-cost and abundant taxi service, car/motorbike ownership is increasing rapidly. You can't beat the privately owned car at its own game.