What does airline competition tell us about robotaxi competition?

A couple of years ago I released my list of factors by which robotaxi companies might compete. Many people wonder if there will be a natural monopoly, limiting us to one or two companies per city, or if we might get more.

I don't think there will be much competition for the first several years. Companies will take over the world one city at a time. Nobody has the management bandwidth to deploy in every city in the USA, Europe or Asia all at once, or even in just a few years -- and few have the money to do that. So if a company wants to deploy in a new city, and they can choose to compete in a city that's already got another player, or go to a virgin city and completely own the business -- which will they choose most of the time?

Still, in some places they will compete, and eventually they will compete everywhere.

Air travel is of course very different from robotaxi travel. Trips are very long, and there are usually no alternatives but air travel. There's 2 hours of wasted time on the ground, if not more. Space is at a super premium and the cost is high. It's almost always public transportation, private jets are for the elite. But it's still transportation, and a reasonably competitive market on many routes.

Clearly with air travel the biggest differentiating factors are schedule and price. Prices are incredibly variable -- sometimes one flight can be 3x or 4x as much as another, and there are many types of passengers (tourists, business, etc.) with very different needs and budgets. But in those cases where two airlines have roughly similar offerings in terms of schedule, what makes the competitive edge?


Most planes have 2 or 3 different seat sizes, with radically different prices. On long-haul, the business class seats are where most of the airline's profits come from, and the coach passengers are just there to fill the part of the plane (at lower profit) that couldn't be filled with biz class passengers. There is surprising variation in the form and comfort of business class seats, and regular evolution as well. Of course, these are seats you will spend 8-12 hours in, and want to sleep in. People care a lot less on short-haul flights, and people who would never book a long-haul coach flight will take an hour flight that way.

Customer service

Airlines work (or don't work) very hard on the non-flight aspect of their service. While the central aspect of a business class seat is how many square feet it takes up, they win customers by trying or failing to treat them well on the phone, on the web, in their app, in the terminal and on the ground. A big differentiator can be how well they treat you when something goes wrong. This suggests that even if robotaxi rides are seen as commodity travel from A to B, the win may come not from the ride but how passengers are treated.

In the case of Uber vs. Lyft, where the drivers are the "customers" from the point of view of the App, Lyft took the approach of working harder on good driver relations, believing that would lead to more drivers and better treatment of customers.


Airlines are the most extensive users of loyalty programs. It's ridiculous compared to other industries. I've done silly things to maintain my 100,000 mile status on airlines, because they treat me a lot better when I do. Many people do "mileage runs" where they deliberately take a quick-turnaround round-trip for no other reason than to get loyalty miles. Loyal customers are rewarded with discounts, free trips, upgrades and most of all better service. It works, especially for business flyers who are spending company money, not their own.

Uber and Lyft have loyalty programs for their drivers, giving them better rates the more they devote to one company. They only do limited loyalty offerings for passengers but this may change. Uber gets an automatic benefit from its size -- when I go to a foreign city, it's much easier for me to just use my existing Uber account than to download and set up an account on the local TNC. The local TNC has to have a lot of advantages over Uber to win the business. The largest robotaxi company will get the big share of the tourist/visitor business. Loyalty programs will make that even stronger.

Food and Entertainment

While people write a lot about the varying quality of food and in-flight entertainment systems, I don't think many people select their airline based on these. Of course, food is not likely to come with short robotaxi rides, and they may not even want you to eat inside at all.

When it comes to entertainment, we all know that most in-flight-entertainment (IFE) systems are dreadful. Turns out there's a reason: Screens in headsets must be tested for crash safety -- you could smash your head one them -- and that's expensive, so systems are rarely updated.

Many airlines have stopped putting IFE into short-haul planes, expecting passengers will just use their own device. In some cases, they hand out loaner tablets. This is even happening on long haul flights. I often use my own phone rather than the IFE because I get to pick the movies, and it doesn't pause when they announce that duty free shopping is now open in 3 languages.

I've seen many companies talk about their plans for the "in car experience" including entertainment. Except for tourists, who want a tour guide, I suspect there will be no "in car experience" other than having your phone drive the in-car screen and speakers.

Regulatory support and competition

Airlines work to use and abuse regulations all the time. They get monopolies within their country, they get control of gates at airports. The airline that keeps the best relations with the government is sometimes the airline that wins.

The usuals: Brand, marketing, etc.

Like most companies, airlines compete using marketing and brand. At present, TNCs like Uber and Lyft do very little marketing to consumers, but they definitely have strong brands and reputations (good and bad.) Because TNCs don't own their cars, branding on the cars is minor (a small sign, or in early days for Lyft, a mustache on the front of the car.) Robotaxis may have a much more obvious livery to build brand, and they will also have their distinctive sensors and designs.

Complex pricing models

Airlines compete on price, and have perhaps the most complex pricing models of any consumer service. The price of an airline seat will vary immensely during its sale period. Airlines build complex models to predict demand and load as well as customer tolerance for prices. They place complex rules on use of tickets to create different price classes.

Taxis have government regulated pricing. TNCs started with fairly simple pricing, with the addition of "surge" pricing at busy times. More recently, they have moved to taking the destination and pre-calculating a price, with much less transparency into pricing.

Prices on airlines, as the photo above shows, often vary greatly to different locations. It is not uncommon for a flight to one destination is much cheaper than a flight to the intermediate city you change planes at.

Might robotaxi rides get complex pricing models? They are not normally bought in advance like airline rides. We hope that riders will be able to comparison shop based on price. My personal experience is I try one company, and if the price is normal, I use it, but if it's surging, I check the other company. Right now, TNC companies resist letting you use a "shopping engine" which will compare prices for you. Often the big player will refuse to participate in the shopping engine, but the smaller players will band together and use it.

Any other lessons from airlines which might make their way into the robotaxi world? Most other forms of transportation (other than car rental) have very simple pricing.

I have updated my article on competition to include some of these factors.


Since I can buy an SDC and share it with people I trust (or will be able to soon enough), that creates an effective maximum price which robotaxis will have to beat. And if their pricing model is too complicated, people will simply resume private ownership.

So I'm not too worried. Things will definitely get better. Maybe a bit more confusing for a while, but that's ok.

Buying your own car will be only modestly more expensive in money than hiring one as needed. Perhaps about $5 to $10/day more as things mature. Presuming you drive it a reasonable amount. If what you buy and hire are the same, which they won't be.

Your bought car might want to be like today's cars -- 4-5 seats, etc. The care you hire for most of your trips won't be, it will probably be a small cross city car that costs half as much per mile to operate.

And if you decide to buy and share a car, you will share that $5-$10 per day with others -- but of course when more than one person wants it, you'll need to hire a taxi during those periods.

I doubt that buying a car will be more expensive than hiring one as needed, for all but the least frequent car users. Remember that prices are generally set based on marginal costs, not average costs, and the marginal cost of using someone else's car is unlikely to be less than the average cost of using your own car.

Look at Uber's rates. Here they are $1.11 base fee, $2.35 service fee, $0.13/minute, and $0.96/mile. Plus a tip. Yes, a lot of that fee is due to the need to pay a driver, but subtract a reasonable fee for the driver and you're still over $0.80/mile, which is much more than the average driver pays for the vehicle they own.

With Uber, which averages about $2/mile, 50 cents of that goes to Uber and is spent on insurance, NRE and driver incentives. Of the part that goes to the driver, there is supposed to be a profit, which is effectively the wages of the driver. In the robotaxi, there won't be special insurance and there will be no incentives or wages for drivers. There will be tons of NRE, but over time that should drop down quite a lot.

Yes, Uber is expensive. That's my point. Cab service is expensive, and most of that expense will not go away just because we have robot drivers.

$1.50/mile is a *lot*.

(I'm not sure you can subtract insurance, either. Insurance costs for running a robocab company will still be significant, especially in no-fault states, or in situations where the passengers don't have insurance themselves to cover being hit by an uninsured or underinsured driver.)

Regular drivers pay an average of 6 cents/mile for their insurance. It doesn't cover when you drive professionally so Uber has to buy insurance for their drivers. I believe they pay more than that, but have not seen the figures. Robocar insurance should be cheaper, even in no-fault. The cars should not be released until their accident rate is lower. However, at first their cost per accident will be higher. In no-fault, having fewer accidents still is a win, just not as much of one. I think no-fault should vanish for robocars in any event. Robocars will all have 3D recordings of the accident, so finding fault will take about 20 seconds. There is no longer any merit to no-fault in that world. No-fault is loved by insurance companies because it reduces the cost of claims if there is no expensive fighting.

I don't necessarily agree, but whether it's $0.80/mile or $1.50/mile or $0.74/mile, it's still a lot more than most people pay for their personal vehicle.

By the time that there are transportation as a service (TaaS) companies offering rides in AV's, there will also be 'nearly' full self driving upgrades for most newer cars. For example, with the dongle and processor you get from comma.ai , you can take advantage of your newer Honda, Toyota, or Hyundais sensor systems which already include cameras and radar. A large number of cars being produced today already have lane keep assist and other safety features which require a sensor suite. There are already companies that let you 'hack' your sensors to allow self driving, so there should be competitive pressure on TaaS companies prices.

Also, they will still offer cheap trips during low demand times. When they have excess fleet capacity, they only need to sell above the marginal cost of the ride (a bit of electricity and depreciation).

I am skeptical of aftermarket add-ons, as long as there are safety certification requirements, and there almost surely will be. Will you be allowed to "mod" your own car without doing an expensive safety cert? Will the add-on seller be able to certify a car they've never seen in all its variations? Something they never tested in that config? Driver assist does not need safety cert because the driver is in control. Once the car is roaming the streets with no driver, somebody has to take responsibility for its safety.

The question of lower prices off-peak is an interesting one. I predict parking will get so cheap that there is not much incentive to reduce prices to avoid just sitting there. If the car is weaering out by the mile, rather than by the year -- which it will -- this should not be as big a factor. Same for the capital cost of the asset. But it could still happen.

But the marginal cost of the ride is not "a bit of electricity and depreciation." It's mostly depreciation (as it is for cars today) but also maintenance, "insurance" and particularly battery depreciation as well. These are the same off peak and on. On-peak it will be easier to get a shared ride, so it might actually be cheaper to ride then!

Apparently the plan with comma.ai is to just get an insurance cpany to underwrite their software product and then just charge 30-40 per month for licencing. As long as insurance is covered, other laws won't matter so much. Human driven police cars arent going to ticket an empty vehicle that is less dangerous on the road than they are.

The marginal cost will be depreciation plus some electricity. Maintenance will be tiny and insurance will be reasonable in non fucked up legal climates. Depreciation per mile on an EV is a lot less than an ICE. Like I said, at non-peak times, taxis will be near marginal cost (whether you include insurance and maintenance in that figure doesn't change that riding will approach marginal cost at non-peak times) and there will be higher prices at peak times.

Is making an autopilot at present. Should they wish to release a full robocar tool, able to move a car on public roads with no human supervisor, the odds are extremely high that doing so will require much more than insurance. It would be interesting if they didn't, but that's not the way the winds are blowing. However, it may be that the maker just has to self-certify the safety. Regulations cover the sale of a car (and will probably be extended to cars that are not sold but only hired out) and the operation of a car, and to some extent the modification of a car by companies. There might be a loophole for modification of a car by its owner, but I don't know how long that will last.

Definitely agree that prices might be cheaper on-peak, due to it being easier to get a shared ride. In fact, I think this is likely, assuming that you're willing to buy your on-peak passes on a monthly basis (e.g. one ride per weekday between point X and point Y any time between 8AM and 9AM). On-peak and single-use might be higher (although it might also be lower if you can fill some excess capacity), but on-peak and pre-booked (a la a commuter pass) should be cheaper.

During much of the off-peak time, there probably will not be excess capacity, because the average number of passengers per vehicle will be much lower.

During the very-off-peak times (3 AM?) there will be excess capacity. But the cars most likely won't be parked nearby, which greatly increases the costs. Which brings me an idea. What if people could let robocar companies use their driveways for parking in exchange for cheaper off-peak rates? I wonder if the local regulators would find a way to do away with that.

I think the cars will be close by at night. The roads will be lined with them -- since you don't need road capacity at night, might as well use it to store cars that will wake up and clear those parking spaces when the morning comes.

People will ride together at rush hour, but don't overestimate how many. Today, our roads are only a little over capacity, with most commuters solo driving. There are many places where, if there are no accidents or other problems, traffic flows fine. Taking out 20% of the solo commuters would do the trick, until induced demand arises. But in metered roads, induced demand can't rise because you don't allow more cars to try to use the road.

I get the impression you must live in a major city. I highly doubt the government will let robot cars park in travel lanes at night here in suburbia. I'm not even sure they'll be allowed to park on the streets at all in residential areas, given that they will be considered commercial vehicles. Parking them in residential driveways would be an excellent solution around here, both at night and during the lower demand parts of the day. I'm not sure if that would break any county ordinances, though (whereas I am fairly sure parking them on the streets in residential areas would).

Whether or not many more people will be willing to ride together at rush hour, I don't know. But there will be a high demand during rush hour just from people switching from traditional mass transit to shared taxis, if the price of a shared taxi comes down. And I don't really look at it from the standpoint of capacity (that's not the benefit), but if I did, note that the people switching from buses (or other traditional mass transit) to four-seater robotaxis will act to increase demand for capacity.

I doubt very many people who aren't willing to ride with couple neighbors at rush hour will be switching to robotaxis for their daily commute. Maybe in major cities where parking is expensive. But otherwise, it'll likely be too expensive compared to using their personal vehicle.

Any robotaxi company is definitely going to need a bunch of multi-seat vehicles, as people often like/need to travel in couples or small groups. During rush hour the overall demand will be high, but the demand for multi-seat vehicles will likely be lower, so they'll have an excess of multi-seat vehicles. If they have a vehicle which can seat four comfortably (six less comfortably), it'd make sense for them to charge two to three times as much to someone who insisted on riding alone, compared to someone willing to buy a monthly commuter pass and willing to share with up to three neighbors going to similar work locations.

Maybe the robotaxi company will also have some single-seater vehicles, and maybe someone with a monthly commuter pass might get lucky some days and get to use them. But people who insist on using them will pay significantly more than people willing to share and pay for a monthly commuting pass, and given the marginal costs it'd make sense to charge at least two to three times as much.

Off-peak, it's less clear what opportunities there will be to save money by carpooling. Some areas will have it, at least at some times. Some probably won't.

I dunno, maybe I'm wrong, but as I see it, a shared taxi commuter pass will cost a little more than a bus commuter pass, and will be much much more convenient. Everything else will cost at least two to three times as much, not a whole lot less than Uber costs today.

Here in Silicon Valley suburbia, the main drag, El Camino Real, has 3 lanes plus a lane of parked cars each direction. You could easily drop it to 2 lanes of road and 2 of parked cars, or even on lane of road and 3 of parked cars, at night. And yes, robots don't park, they stand, so they can double park or park in front of driveways and fire hydrants. Plenty of places for a car to wait near everybody at night. Even if there are not, you only need a few spaces, because as soon as one car is summoned from close by, another car can move in from further away.

I forecast the robotaxi ride to be cheaper than your own vehicle. So why not switch?

Fleets will have multi-seat vehicles, but it's flexible. You can serve one person in a 4 seater (that's what we do today) but it is not as efficient. You can also server 2 people with two single seaters, unless they insist on a social trip. But if that's all that is available, they can chat on their videophones.

Even off peak, sharing will be cheaper for trips of 20 minutes or more. If we imagine the single seater is 30 cents/mile, and the 4 seater is 50 cents/mile, if you are doing the 100 mile round trip I did last night to a party, then the solo price is $30, and the share with 4 price is $12.50. If the solo vehicle pays half the road tax the difference is greater.

While so far the efforts to encourage ride sharing have had minimal effect for a large investment, that has hardly ever discouraged regulators from trying again.
Authorities have an enormous amount to gain from increasing load factor, and providers with better coverage and larger fleet will more easily meet this demand.
The form the incentive takes might be monetary, but if as you say rides will be cheap, the the incentive would likely be in the form of a tax for single occupant rides or congestion charging for certain zones, or both.
Some cities might decide to allow only a single operator and reward/punish them on the basis of that KPI.

I think they will. They will probably botch it, trying to pick winners and methods.

The right way to encourage efficient travel is to charge a fair price for road usage, a price that is higher when it gets congested. Charge it to everything, including transit vehicles, which should get no subsidy that everything else doesn't get.

Such a charge would make people pool together if that's needed. Since it would be congestion-based, the charge should get very, very high at the peak of rush hour, making it quite costly to go solo at that time, until enough people stop that the charge goes down.

Hmm, I wonder if the robotaxis could be used to transport packages between hubs at night. I guess it's a matter of whether or not you could stuff enough packages in one to pay for the gas and wear and tear and insurance. It seems you could, though. They could go round-trip overnight, or you could send enough in the opposite direction and do a switcheroo each night.

Much thinking about cars depends on the old idea that they wear out by the year, and that when not in use they are an "idle" asset that is wasted. That is much less true when they are taxis and wear out by the mile. The only thing wasted is parking and the small interest on the capital. Both are quite low, so to use a vehicle for other than its designed purpose is wrong if it's not efficient at that purpose. So if a cargo robot is more efficient, use it even if a passenger robot is sitting idle.

What's wasted is the opportunity cost. If you can earn enough money transporting packages to pay for your expenses plus a profit, you should do it.

Actually, nevermind. I get it. If you can buy a vehicle dedicated to transporting packages for $75,000, and run it 200,000 miles a year for 10 years, the cost of the purchase of the vehicle itself is only $0.04/mile.

Still, it might be something that could be done during the peak package season, when your vehicles dedicated to transporting packages are all full. :)

Well, at least with today's vehicles, you can't run a vehicle 200,000 miles/year for 10 years. That's almost 600 miles/day which is almost an impossible duty cycle except for something like a inter-city shuttle. But there is no vehicle with an expected life of 2M miles. Most US cars last 200,000 miles. Electric cars might do better.

The whole vehicle is worthless after 200,000 miles, even if it's only a year old?

Is the LIDAR sensor broken after 200,000 miles, regardless of the amount of time?

More likely the LIDAR sensor is worthless (or, at the least, worth *much* less) after a certain number of *years*, when new technologies render the old technologies obsolete.

Not the whole vehicle, but as close as they can get to it. Today, you try to engineer a vehicle for a planned lifetime. You design everything so it will last the expected life, and not much longer. Of course, things don't degrade as a step function, and the lifetimes are not exact. Primary things to wear out are moving parts (which are fewer in electric cars.) Batteries and high-power electronics. Seats and interiors that are worn with use. Exteriors that suffer wear from small things thrown up by the road, rain and snow. Frames from flexing. Brakes from braking. Things that heat and cool repeatedly.

Other things won't wear out this way, but the problem is the cost of repair right now is much higher than the cost of manufacture. Most cars die not because all the parts are worn out, but because repairs are going up in frequency. Due to labour, replacing a motor or water pump or timing chain can cost you thousands, and once your car's book value is only thousands, repair is no longer economical. You junk it, and then in one go they extract all the parts that still have value, but mostly to use in repairs of other worn vehicles as nobody wants to put used parts in new vehicles.

Taxis go about 250K miles in 5 years and are pulled from service.

The move to electric changes a number of these things, in ways we have yet to learn. And as you suggest, there is also things going tech obsolete. Your 2 year old cell phone works just fine, except maybe the battery. Nobody wants it. A 5 year old LIDAR may still perform to spec. But the newer ones have much better spec and meet new safety requirements.

FWIW, my comments above were about a long-haul package delivery vehicle. 200,000 miles a year is, admittedly, about double what you might expect from a long-haul truck, but I figured the lack of a human driver might make that reasonable. In any case, I would think you would want to design a long-haul package shipment vehicle with replaceable parts if the parts weren't expected to last more than a year under your planned driving conditions. Replacing an engine isn't going to be cheap, but it's not unheard of in the long-haul shipping industry (and would be even more common if it weren't for the fact that these things are built to last more like a million miles than a mere 200,000). Battery life might be even shorter, but those are even more readily replaceable. You generally junk the whole vehicle (or sell it to someone more desperate) when a new, more efficient one is available, which is a function of the number of years old the vehicle is. If it's not that old, you're more likely to repair the failed part.

Maybe that's different in the cab industry, but it sounds like it isnt. "Taxis go about 250K miles in 5 years and are pulled from service." The question is, if you drove the taxi half as much, would it really have half of its value left after 5 years?

Probably not. After 5 years, there are probably significantly more efficient (and safer, and more luxurious) vehicles out there, and at some point you're probably better off selling the vehicle for 1/4 (or whatever) of its original value and buying a new one. So time is an issue, and there is a cost (in depreciation, not just interest) in having it sit idle.

That said, I agree I probably went way overboard with my musings about using robotaxis to deliver packages. :)

Opportunity cost only exists when there are willing customers (waiting opportunities) and no way to serve them. What I am saying is when your vehicle wears out by the mile, there is no difference. Your floor price is the same at all times.

Think of your car like a tank full of gasoline. You could drive somewhere and burn up a gallon, or not drive and keep the gallon. Much more like a consumable entity than a fixed asset.

The market will provide enough transportation for the public's demand. It will actually provide somewhat more.

Opportunity cost exists when there are willing customers and no way to serve them more cheaply.

But I guess I take issue with your assertion that a vehicle goes down in value solely by the mile. A vehicle bought today is not going to be worth the same amount in 5 or 10 years, even if you don't drive it at all. And not everything on a vehicle even "wears out" by the mile, especially if it's parked outdoors exposed to the elements.

It is indeed a complex formula. What I point out is that for taxis, it's much more "by the mile" and for owned cars it is much more on the "by the year" access.

In many cases, it involves conscious plans. If the loss of finish that comes from parking outside can be quantified, then you can decide if it is worth the extra cost of driving to protected parking or not. If you want the finish to last decades you might. Or you might factor in the cost of repainting.

This is a complex issue and nobody really knows the final values. But we learn something from existing taxis and their life-cycles.

One of the big questions you will look at is "How many vehicles do I put in the fleet?"

More vehicles requires more capital cost and parking cost. But they get lower use so they last more years. More vehicles means more idle vehicles to serve customers, which means shorter wait times for them, which is an important competitive edge, worth the money up to a certain point. But make the vehicles last too long and customers don't want to ride in them near the end of their life cycle.

I agree completely.

Add new comment