How do we get robotaxi rides down to 30 cents/mile while Uber/Lyft charge 50 cents/mile for just connecting you?


In discussion of the eventual cost of a robotaxi ride, I and others have forecast costs similar to the all-in cost of car ownership. Today that's 40 to 60 cents/mile (plus parking) and for a one person electric "city car" it can be under 20 cents. Note that in building these costs I am looking at the full retail cost today including:

  • All profit margins
  • All marketing budgets of existing car, banking and energy companies

As such, it makes sense as a retail price for somebody who is selling you the whole enchilada -- a ride from A to B with everything included. You need to add a bit for empty vehicle moves (about 10-15% of miles,) and the cost of parking the car when not in motion -- but that parking is going to be cheap because cars don't need to park exactly where you want to be dropped off.

One big number contradicts all this, however. Today, services like Uber and Lyft sell rides for about $2/mile. (I see a range from $1.50/mile on long rides to $2.50 or more on shorter rides. Super short rides have a $5-$7 minimum.) While they started at 20%, today they take 25% of this money for their own fees. In other words 50 cents/mile. And for that, the value they provide is the largely the app that connects riders with drivers.

Where does that more than 50 cents/mile go?

  • Some of it goes to insurance for the trip, which really does cost by the mile
  • About 2% (4 cents/mile) goes to credit card companies, but transport is so big that your transport company might bill you directly with cheaper means
  • A modest amount goes into marketing. These companies don't market much to riders, but they do market to drivers
  • A large fraction is going to incentives to drivers to recruit and keep them
  • The rest mostly goes to paying their staff in a fast-growing business. Maintaining software and many other things. A bit for servers and operations.
  • None of it goes to profit -- in fact, they are losing money at this price, some say losing it fast.

Yet they do a tiny part of the actual process. The cost of simply arranging a ride should should dwindle to almost nothing with time. It should become a flat fee per trip, not counting insurance, because none of their costs depend very much on the length of the trip -- other than you can do fewer long trips in a day than short ones, so more fixed cost must be allocated to the long trips.

As noted, they are in growth mode, so a lot of their revenue -- and their losses -- are spent on this growth. In particular, incentives to recruit drivers.

The robotaxi company doesn't have to recruit drivers or keep them happy. It just has to get capital to buy cars and pay the interest on that capital. That's not a lot at today's interest rates -- around $1/day on that $10,000 single person car. And car ownership actually has a higher capital cost because the car sits idle much more of the time.

A robotaxi company needs an app, of course, and a fleet scheduling and positioning system. It needs to take requests from customers, and dispatch cars to them. It needs to do strategic repositioning of cars to lower wait times. As does Uber, though it puts the cost of this mostly on drivers, who are not paid when driving to you or to a surge zone.

Over time, these companies will cut their costs. Will they cut their price to drivers? A competitor trying to break into their market today would probably be very attractive to drivers if it charged only a flat fee per ride, rather than this percentage. But nobody has had much success in competition without taking the big money and big losses approach.

But it will happen. And to get our robotaxis down to the super low prices that are possible, it has to happen. But it won't happen on day one, and as such early robotaxi services will probably cost $1/mile.

It will also be interesting to look at the cost of the hybrid robotaxi services Lyft and Uber will offer. These services will serve you with a robotaxi if your trip is a route the cars can drive. If you want a different route, they will serve you with a human. They can either charge $1/mile for the first route and $2/mile for the second, or perhaps can charge some price in between for all rides, subsidizing the human driven rides with the robot ones. This strategy makes sense when you are looking at subscription services, and services for rides other than your standard commute. Nobody will offer you a robot-priced subscription service unless your most common rides (ie. home to work) are in the robocar's territory, and human drivers are only needed for your more random other trips.


"But it won't happen on day one, and as such early robotaxi services will probably cost $1/mile."

John Krafcik talks about waymo pricing initially. 31:54 he mentions it'll be similar to uber and lyft, maybe cheaper? So probably twice what you are estimating.

While Google does not need money, in the early period of operation, the novelty of robotaxi service will make people who are willing to ride in it quite keen to ride in it, so they would indeed pay a similar price to Uber. I had not been predicting they would not try to have some price edge, just to teach people about how it's going to work, because it's not any more world changing than Uber is to be priced the same as Uber.

This might also reduce complaints about job displacement if the robot "charges" as much as a human.

The reality is that in early operation, it will cost them more than an Uber to run because it's new.

Even flat fee per year. The cost of building the app (which I use to include the backend for the app) doesn't depend very much on how many trips you arrange.

Might be an interesting model to try, though ultimately what pricing scheme is best depends more on the demand curve than the supply curve. Probably a high price per mile with a monthly/annual subscription plan that drastically lowers or eliminates that per mile rate will better capture profits from traditional cab users - don't need the service often but are willing to pay a lot when they do - while keeping the service affordable for those who want to use it regularly.

However, I think if an Uber competitor were to come out and say, "Drive for us, we charge only $1 per ride or 5% of your take" drivers would flock to that. They don't love Uber, and they like Lyft better but not that much.

No driver marketing or recruitment needed. But nobody has done this, even with a flood of VC money.

I bet if an Uber competitor charged nothing they'd get even more drivers than that. Perhaps VCs aren't eager to invest in a copycat business that's poised to lose even more money than the company they're copying off of, though.

$1 or 5% probably is barely to pay just for payment processing and dispute resolution. eBay + Paypal charges 8+3% to 15+3% for an item that costs $50 or less.

And all of that for a business model which is guaranteed to be obsolete. The robot version of Uber is exactly the same as the robot version of any other taxi company. Uber's main innovation was in evading taxi and employment regulations, and by replacing the employees with robots and billing customers directly, they'll lose both benefits.

Yes, you would lose money at 5% for the reason you say -- and one must add the insurance which needs special coverage for human drivers (the robots would be insured for all use another way.) However, when the world transitions to robotaxi rides (which is not overnight) being Uber is still great even if you don't have your own internally developed robotaxi. Tons of companies are itching to make them to sell to you. If, like Uber does today, you can gain a major brand in selling rides, you are in a great place to be. So if there is a strategy which would seduce all the drivers, VCs would be willing to spend money on it.

is how much the cost per mile of car ownership will go down. A family of three (two parents and a teenager) might be able to get away with one car instead of three. Additionally, car insurance costs will probably go down drastically.

Depreciation costs will likely also go down as the cost of a car becomes mostly software that can be updated over the air.

Yes, for users, the cost of owning cars will go down if they cut back how many cars they have in the house, and for various other reasons.

When cars are computers, I don't think it reduces depreciation -- in fact, it may make it worse. Look at the depreciation on a computer! For taxis, depreciation will be mostly by the mile, not the year any more.

I didnt, and wouldn't, say a car will be a computer. If you really want to go with that analogy, different parts of a computer depreciate at different rates.

The cost of the *software* is going to be a key factor, I think. Hopefully there will be an open source option. A car with open source software and parts that are as modular as a desktop PC would be a great buy that could last quite a long time.

A car is not strictly a computer, but the most important part of the car will be the digital systems that drive it. If they are 5 years old they will be obsolete. You need to design it to be cheap to replace them so that the parts that aren't a computer can be re-used. With many consumer devices we throw out the parts that have not gotten obsolete because the computer inside has.

It does not look good for open source. It is likely any car that gets on the road will need to be certified by some party as safe and go through some level of government approval process. There would need to be some sort of exception to that for a tinkerer's car, put together by its owner and used by its owner. For a taxi fleet, you would need a big entity which does the certification of the open source code.

Open source is not synonymous with "tinkerer."

In any case, the advantage if open source would be to protect people from the company going out of business. I'd strongly hesitate before buying a self-driving Tesla because of that fear. If Tesla open sources their software...maybe I'd buy one.

I mention tinkerers because there is a long history of automotive tinkering (or the original good meaning of hacking as we would call it in the computer world) and that history gets respect. The founders of all the car companies were tinkerers, back to Karl Benz.

The fraction of people who refuse to buy non-open-source tech is unfortunately insignificant to the industry. And you aren't going to buy these cars, you are going to ride in them. There is no worry about the company going out of business when you are just riding in it for 15 minutes, unless you think it is going out of business in the next 15 minutes.

I hope that we have more sense than to give up on the idea of personal ownership of a car. Here in the USA, we might. Having your own car, or at least sharing in a family car, and the freedom and privacy that comes with it, is an essential part of the American way.

This is often said, but then you are saying people who live in Manhattan, who give up on car ownership immediately upon moving there, are not American. Truth is, we are much less wed to that than people imagine, and people just want to get around in the way that makes sense where they live. However, there will indeed still be car ownership very commonly in rural areas, which is a lot of the country, and still some of it in cities, of course.

However, the other point remains. The fraction of those buyers who declare, "I will not buy a car that is not open source" will probably be similar to the fraction of people who won't buy computers and phones unless they can get the source code.

Okay, "essential" was an overstatement. Still, the vast majority of Americans have no desire to live the Manhattan lifestyle. Especially not without the benefits that are somewhat unique to Manhattan. Furthermore, robocars will facilitate personal ownership of cars even in places like Manhattan. Probably the biggest problem with owning a car while living in Manhattan is the parking. Robocars can be parked far away from the dropoff location. Hell, you can park one in New Jersey and have it available for whenever you need to leave the bubble of Manhattan. Some people already do something similar, but they're limited to parking in places near a PATH station or something.

Maybe most millennials will adopt it. They seem to be willing to make sacrifices for the sake of conservation (whether the numbers add up or not) more than the rest of us. But I'm not convinced that most Americans will give up personal car ownership very easily. And this post has helped me see that it's probably not going to be a huge cost savings to switch to the communal model of car ownership. Which I guess makes sense. The increased cost of every family unit having a robocar vs. some taxi company having enough robocars for everyone is not that big. And there's always going to be a lot of overhead involved in doing things the taxi way. Taxis will still have their place, and that area will probably increase. A family of four might cut down from two or three cars to only one, and use taxis for those situations where multiple family members want to travel to very different places at the exact same time. But really, after reading this post, I've begun to think that personal (at least family) car ownership is going to be around for a very long time. Which is great. I'm not eager to switch to the communal car ownership world.

Yes, as I have written, all models will be popular -- ownership, non-ownership, mixed use. And yes, I think the first adoption will be in the 15-30 crowd, as well as the over 70 crowd (not by choice, but because they can't drive any more.)

Rural people will also mostly stick to ownership, and that's over 30% of the market. If you have mistaken me for one of the people who says car ownership goes away, you were incorrect.

Okay. I must have misunderstood you when you said "you aren't going to buy these cars, you are going to ride in them."

Robo cars will need to be parked where the riders are, if not there is a cost to move them in and out of the area in between rush hoir periods. Plus there is no ROI when the car is idle. There is a lot of potential for robo cars in big cities but there is a lot to work out before the potential can be realized.

The good news is that while there are certainly peaks at rush hour, there is still lots of traffic at other hours, though not so much at night. So many of the cars will be at work (one reason why you want to avoid too many big vehicles as they can't be used off-peak efficiently.)

The other nice thing is that there is a natural balance between the road capacity you need and road parking. At rush hour, no cars parked, and you need all the road space. Outside rush hour, cars can park on the streets, because you don't need all of them any more. And in the remaining parking lots. Robocars can double or triple park, double park in front of driveways and hydrants etc. because they are not parked, they are standing. In parking lots they can park more densely than any valet can do it.

And yes, the ones that only work at rush hour will wait somewhere, and it might be a mile or two outside the dense cores. That's not a big deal. For the return rush hour, they can wake up and be ready -- either going into the core, or waiting for people to take quick hops on shuttles that run back and forth between the core and the satellite zones where the cars are waiting. Every car going into the core can be told to come back full and make a quick stop at a satellite transfer point if it wants the privilege of going into the core. But the full vans will not stop, they will head to the area where all the people on them live, and then drop them where a group of smaller vehicles are waiting to do the last mile.

Something else that is going to have to be solved by the robotaxi companies is the use of a car to temporarily store things. Consider a trip to the water park. I'm not going to bring the towels and the sunscreen (or even my wallet, probably) into the water park with me. I'm going to leave these things in the car. Sure, the robotaxi might have some way of taking the stuff to a storage unit and then bringing it back, but that's going to add cost (though probably not as much cost as the $10 that the waterpark charges to rent one of their lockers).

On a longer trip there might be a whole host of items that get left in the car, as I go from place to place. Often things will be bought at one store and stored in the car while you go into another store (or into a restaurant to eat). Again, maybe the robotaxi can offer a service for that, but that's going to have a cost. Probably for a road trip you'd want to just rent a robotaxi for the length of the road trip, if you don't have a car of your own. Which I guess is acceptable, but there's a lot of overhead having to rent a car for a weekend or whatever.

Babies need special seats. These can be provided by the robotaxi service, but that's a significant overhead, and it'll be a pain to always have to adjust the seat for the baby. Furthermore, a baby seat used by strangers' babies should probably be cleaned and disinfected between every use - something you wouldn't do between *every* use if you know whose baby slobber is all over the place. This might be another situation where a short-term rental (for the time away from the house) makes more sense than a pure robotaxi.

Children, in addition to car seats (which are fairly generic, small, and less messy than baby seats) often have toys or books or electronics they use in the car. These things are often left in the car when going to the supermarket, or a restaurant, or the movies, or whatever. Probably in this case the best solution is just to deal with the whining and tell them they can't have whatever it is, and will have to settle for whatever entertainment system is provided in the robotaxi.

I think communal car ownership is a long long way away for most Americans.

In my semi-fiction A week of robocars from 10 years ago, I outline a hypothetical business called the "Stuff Locker" where you maintain a standarized locker of your things, and when summoning a car with a little warning, it goes there to pick up your locker before picking you up. It could solve part of this problem.

Yes, you will be able to keep your car during the day if you are doing multiple errands. The largest cost for that will be parking (just as if you owned a car) because the cost to rent a non-moving robocar does not need to be very large. The capital cost on a small robocar is just $1-$2 per day at today's interest rates, so the main cost is really the opportunity cost of taking it out of service so it's not out making money for its owner. Still, it's entirely possible the cost to keep it will be cheaper than today's cost of parking an owned car, perhaps in the range of $1/$2 per hour. Today we have "free" parking subsidized by cities and businesses. I think that might end but be replaced by "super cheap" parking in the range of 50 cents/hour or less.

You don't need to change the whole baby seat, changing a liner is probably sufficient. I do think parents of small children will be more inclined to ownership for this reason. Ownership is still something many will choose, that's hardly going away. But many people will switch. And the families will switch from 2 cars down to 1.

For the children who aren't constantly throwing up, you will be able to summon a car with built-in seats designed for children of various sizes. Your app will know it's you and your 3 year old, and ask for a suitable car for that, which might be a car with a special toddler seat (facing daddy, who is not driving) or a car that's had something put in it for you. The key is that most travel with small children is semi-planned. You don't know the exact time you are leaving but you have a rough idea. If there is warning, there is time for a specialized car to come to you with no wait when the time comes. When coming to your house, of course you can have any extra things you need at the house.

In spite of all this, though, there are many places (like Manhattan) where car ownership is very low, and somehow they make it work.

> The fraction of those buyers who declare, "I will not buy a car that is not open source" will probably be similar to the fraction of people who won't buy computers and phones unless they can get the source code.

That fraction is high enough to have created Linux and Android. Hopefully there will be something at least as open for cars. I suspect there will be, but maybe not until several years after consumer robocars with proprietary software become available to the public. It's always much harder to be the first than it is to make a copy-cat.

Linux and Android were not created because people refused to buy devices unless they were open source. For Linux, people just bought commodity hardware (sometimes even paying for Windows on it until Microsoft was forced to stop that) and put Linux on it. Android's a different story, but in most cases vendors cut a deal with Google to get the whole Google package on top of the free OS rather than just releasing a purely free software phone.

"refused to buy devices unless they were open source" was your strawman. Most people, including myself, aren't nearly that strict about it.

Many things will be different once RC start to ride in huge amounts. I will point one now, then others later.
Human driving cars (HDC) are not safe. If we have an area with 10.000 cars and 20 accidents per day, with 5.000 HDC and 5.000 RC it suppose will be 10 acc. between HDC, PLUS the accidents HDC cause to RC. RC normally, will cause no accidents between them or to pedestrians or HDC
So rules and test will be tighten for drivers, so less people will be allow to drive. Drivers over 60 years old may be will be banned. People will be reluctant to take a ride in a HDC.
So, soon, not taxi service based on HDC will be allow. (even people will not want to take a HDC taxi). It will not be fight between RC and taxi drivers, just people will not want drivers.
Less people capable to drive, everyday less people who wants to ride in HDC, will push to find the way to have enough RC in the area, and withdraw all HDC. In the first moment the solution will be to oblied any RC in the area to be available to anybody who needs, when the owner is not riding it. Carpool will be also mandatory.

I had not read the blog, I did it now. It is from 2014, but the arguments are still valid. The difference is that we are now more sure of the security of the RC, and its massive circulation is more imminent. Regarding the issue, the opinions are divided. For reasons of time should delay a couple of days maybe, to continue with the subject. I apologize Thank you.

Might not have to outright ban it. Prices of mandatory insurance will go up and up until human driving is too expensive.

The idea of insurance going up (which I have seen many people say) makes no sense at all. Insurance will go down for human drivers, not up, unless some law forces it to go up. The insurance price is based on the number of accidents your class of driver has. Why would the rate of human accidents go up in a world of robocars? It should go down -- especially since the human driven cars get more collision avoidance features.

It could go up because there will be fewer people paying for insurance.

That said, maybe it will go up for some classes of drivers and go down for other classes of drivers.

No, fewer people paying does not increase the price of insurance. I mean if it were a super tiny number the insurance industry might die, but that's a really small number. Even if all the other policyholders are horrible drivers, what matters is what risk bucket the insurance company puts you in. Unless your risk has (inexplicably) gone up, your insurance does not go up. Just because everybody else on the road is safer doesn't make you more dangerous in an absolute sense, only a relative one. Today an average driver pays about 6 cents/mile for risk. If the robocars pay 1 cent/mile, that makes their ride cheaper and more attractive, but your insurance still costs 6 cents/mile. For the very cost conscious, they might be more bothered by the cost -- why am I paying $700 per year for liability insurance when those robocar riders aren't?

The ability of the insurance companies to put people in the right risk buckets is limited. In practice, the ability to assess risk is very limited, and car insurance rates are regulated by governments. Furthermore, different groups have different demand elasticities. As demand goes down, demand elasticity will go up, and that likely will mean prices will go up (remember, it's a highly regulated industry, not a perfectly free market).

It's certainly possible that car insurance costs will go up. I'm not saying it's certain, though. How government regulators respond will play a large role, among other things.

It is regulated, in terms of what factors they can use to differentiate between customers, but the price is not regulated, not in most places. It could not possibly be regulated. A large fraction of the insurance companies in the USA are mutual insurance companies. If they make extra profit they distribute it back to policyholders. The premiums really do have to, in aggregate, match the cost of the accidents plus the expenses of the insurance company. The government can't order it to be otherwise (unless it is government run insurance.)

But don't forget, at the same time as everybody is moving to robocars, the regular human driven cars will keep getting safer and safer. They will get better, cheaper ADAS functions -- collision avoidance, lanekeeping etc. In fact, many of them will be cars that can almost drive themselves and can keep you out of an accident. Insuring them does get cheaper. (In California, you can't use what type of car the person has as more than the 4th highest weighting factor on insurance, so that forces all insurance to get cheaper, and that can't be changed except by constitutional amendment.)

The older cars will still be around, slowly wearing out. But they don't get more dangerous to drive. In fact, with the more regular behaviour of the robocars, they probably get a little safer. While I have not heard of anybody doing it as yet, eventually robocars will even work to prevent accidents that are the other guy's fault as much as they can. It's even possible that, if the law encourages it, empty robocars might "sacrifice" themselves (or at least their fenders) to reduce the severity of an accident they see happening near them. That's some time in the future though, as it's a risky thing and you need to get sure about it and have a good samaritan style law behind you.

Excellent article Brad, thanks for sharing. However, I wanted to ask how much you had considered governmental response to automation in the economics/costs of a robotaxi service? Taking away the drivers will also take away their wages which are usually spent locally and help stimulate local economies. Robotaxis give an opportunity for huge private corporations to hoover up money from local economies through fares and cream huge profits to be hoarded offshore in the hands of the few (in theory - unless corporations suddenly go against all their historical capitalist behaviours and generously pass all these savings down to the customer). Central governments will have to respond with innovate new taxation on automation in business to ensure that wealth continues to be distributed effectively and that we can all share in the benefits of automation - better cheaper services, more free time etc. Currently, governments are still clueless on this response so there is still uncertainty on how much this new taxation will impact service costs. But it is still likely to impact costs, and should be including in any predictions of future pricing - have you considered it?

Just because it isn't rational doesn't mean it can't happen. It isn't rational. Those human drivers kill more people in their work than anybody else (except doctors) and the idea that we would want to tax something that takes away such jobs seems crazy to me. But taxes have been crazy before.

Add new comment