United’s Revenue Management Problem: The Death of Independence of Demand

Fares, United

The most important theme coming out of United’s investor day last week was that it admits something that everyone else has known for years: it has a real revenue problem. Considering the network and assets it has, it should be making more money. A lot of things are going to fix that issue, including initiatives we’ve already discussed like better scheduling of connecting hubs and the introduction of Basic Economy. But there are behind-the-scenes changes in the revenue management system that are going to yield big improvements as well. Scott Kirby opened up the black box and explained in surprising detail some of the issues United faces. Today I’m going to translate some of that into English.

Alright, so maybe that’s not fair. Scott actually did a pretty good job of talking about the issues in plain-speak, and I’d encourage you to listen to the investor day presentation. Once you get it started, flip to slide 42 and the audio will skip ahead as well. It’s slide 43 that starts to explain what’s going on. Or you can just flip through without commentary here.

This slide sums up the four primary issues:

United Revenue Management Challenges

In short, the overall problem is that United’s revenue management system is nearly 20 years old. Since it was built, a lot of things have changed in the world of pricing/demand as well as in technology. So there are issues that need to be fixed. The last 3 issues here are pretty straightforward, so I’m going to focus on the first one, the independence of demand, or lack thereof.

This sounds like a revolutionary war cry of some sort, but it’s really referring to how pricing structures have changed over time. First, I need to explain how revenue management systems work. Airlines file fares which book into so-called “buckets” which are represented by letters. We’re going to use a very simplified example here to make it easier to explain.

For United, let’s pretend it has only three buckets in coach (there are far more). The coach bucket with the highest fares is Y, and that’s followed by B, and M in descending order. These letters mean nothing. They’re just ways for United to determine how many seats it wants to sell at each fare.

Let’s say there are 100 seats to be sold on a particular flight from, say, Los Angeles to Denver. There are three fares in the market:

Bucket Fare Rules
Y $250 one way
  • Refundable
  • No advance purchase
B $250 roundtrip
  • Non-refundable
  • 7 day advance purchase
  • One night stay required
M $100 roundtrip
  • Non-refundable
  • 21 day advance purchase
  • Saturday night stay required

Demand forecasts show that there will be 25 people who are willing to pay for a Y seat, 50 who will pay for a B seat, and 75 who will pay for an M seat on this particular flight. Since United has only 100 seats to sell, it can’t take everyone. So it will work to maximize revenue, selling as many of the high fare seats as it can.

Buckets are “nested” meaning that anything available in B will also be available in Y. (After all, if more people want to buy the top fare than forecasted, hooray.) That means every seat, all 100, will be available in Y. Then 75 seats will be made available in B (reserving 25 that can only be sold in Y). That means only 25 will be allocated to M class.

These fares are all pretty unique, and the people who are truly price sensitive will only go for the cheapest fare. Meanwhile the business traveler who needs flexibility and can’t stay overnight or book far in advance will buy the most expensive fare. This is what United is calling “independence of demand.”

It’s also roughly how things used to work when Orion was built 20 years ago. It’s NOT how things work anymore. The increase in low cost carrier competition has led to an erosion of fare rules on a much greater scale than back in the 1990s. Now you see a fare structure more like this.

Bucket Fare Rules
Y $250 one way
  • Refundable
  • No advance purchase
B $125 one way
  • Refundable
  • No advance purchase
M $50 one way
  • Non-refundable
  • 7 day advance purchase

Again, this is a highly-simplified model. But now what you find is fewer fences to wall off the different fares. In many markets, roundtrip requirements have disappeared entirely, let alone the Saturday night stay. And instead of having one fare with unique advance purchase requirements, you could have multiple fares with the same rules. How do you determine demand?

The reality is that there is no reason anyone would ever buy the Y fare unless United made it so that the B fare was unavailable. And while there are still differences between the B and M fare, the roundtrip requirement is gone and the advance purchase has shrunk. Overlap of demand for the different fares increases. And this is where Orion stumbles.

Let’s just focus on the Y and B buckets here for this piece. Orion will look at historical demand in each bucket to guess demand for our example flight. Let’s say that Orion is looking at 100 flights in its history for this particular forecast.

If on 50 of those flights (half the time) the B bucket is open, that means demand for the Y fare is zero, even though the Y bucket is also open. With identical fare rules, you would never buy the more expensive Y fare as long as the B fare was available. On the other 50 flights, the B bucket is closed (probably due to manual intervention from an analyst – more than 2/3 of flights are handled this way at United), and demand for the Y fare is 20 people on each of those flights.

Selling 20 Y seats on 50 flights (with none sold on the other 50) means United sold a total of 1,000 Y fares. Divided by the 100 flights and we have an average of 10 per flight. That means Orion will assume there will be demand for 10 Y seats on our flight, because it doesn’t realize there would have been demand for Y fares on the other flights, if only the B bucket had been closed.

And that is the issue with the disappearance of the independence of demand. This forecasting algorithm wasn’t an issue when fares had greater fences separating themselves, but now it is a huge problem. The system underestimates demand for higher fares.

To make it worse, this causes what United is calling “spiral down.” When the system underestimates demand in Y, it holds back fewer seats to be sold in that bucket (10 instead of 20 in our case). And then it underestimates demand off the new, already too low estimate. The system effectively shrinks expected demand for the highest fares closer and closer to zero.

In practice, this means United is holding back fewer seats to be sold at higher fares and leaving lower buckets open for longer. That may be why you can get a cheap fare at the last minute even though you’re willing to pay more. Now United is going to fix this issue (in the next year or two), and that means there will be fewer seats available at the lowest fares and more at the highest fares.

By United’s estimates, these changes (not just fixing the problem of the lack of independence of demand, but all four) will be worth $100m in 2017 and upwards of $900m by 2020. That is, of course, if United’s forecasts are right….

For more reading, check out United’s presentation beginning on page 43 of this deck.

Get Cranky in Your Inbox!

The airline industry moves fast. Sign up and get every Cranky post in your inbox for free.

27 comments on “United’s Revenue Management Problem: The Death of Independence of Demand

  1. The only thing that is mind-boggling about what UA said about its revenue mgmt. systems is that they took so long to recognize the problem and work to get it fixed. As evidenced by the value of the improvement, revenue mgmt. technology easily pays for itself very quickly. Given that there are a number of problems w/ its RM systems, either the former mgmt. regime did not recognize the problem and push the case for fixing or replacing their RM systems or UA senior mgmt. was woefully unable to realize what it takes to make airlines work.
    Given that AA has said they have revenue mgmt. system issues as well, it is really hard to imagine how investors or analysts have been as patient in getting these problems fixed.
    None of these problems have uniquely arisen just in the past couple years. Low cost carrier pricing structures – which legacy carriers have matched with varying degree of vigor for years – are here to stay. If UA is finally admitting that it hasn’t been prepared to compete in a world with abundant low cost competition, the roots of that go back decades and is equally true for AA and DL. If a legacy carrier hasn’t figured out how to successfully compete with low cost carriers,-regardless of the strategy, while preserving the premium revenue which legacy carriers are supposed to be able to attract in order to cover their higher costs, then it isn’t likely that a few changes to a revenue mgmt. system are going to fix the problem.

    More significantly, since revenue mgmt. system issues are something that no investor or analyst could know anything about unless the company says something, you have to wonder about the true motivations in going into such detail. If either AA or UA has revenue mgmt. technology that cannot support the competitive strategies which they have been using, the onus is really on the people who have sat in the big chairs to explain why these issues were left unaddressed for as long as they were.

    Given that analysts and investors can never verify if the airlines in question ever reached their improvement goals as a result of revenue mgmt. technology improvements, you have to wonder if all of this isn’t motivated by a need to throw out some big numbers that SHOULD lead to revenue improvements but which really can’t be verified and could end up being just big fudge factors that might calm down investors and analysts for a few years.

    1. I wonder if part of the motivation for UA to come out in this much detail is based on all the management changes. Smisek gone, bringing over Scott Kirby from American, etc. They can go into detail about the problems and how they intend to fix it, since they don’t have to take the blame for getting the company into the situation in the first place.

    2. I would think like David that this is said from perspective of “outside expert” coming in.

      As to your point about why it can’t quickly be fixed – there aren’t many competent RM firms out there that really understand and can do better. Airlines are still on the forefront so there aren’t many revolutionary solutions available off the shelf. Additionally, as most are demand-forecast based, even if you have a new system, it can take a long time to realize results as you collect data and fine-tune the model.

      1. Airlines the size of UA often have in-house experts on their revenue mgmt. systems. I don’t know if UA has those kinds of people now, but they have had in the past. They are highly educated with mathematical and IT backgrounds and there are people who have those kinds of qualifications. In order to have in-house forecasting engines – which all of the US big 3 have for reasons noted here- airlines must have in-house experts.

        You are correct that fixing the forecasting problems doesn’t lead to immediate revenue benefits because history and forecast validation and correction has to take place; I suspect that is precisely why UA’s revenue benefit is modest in the first years but increases to fairly significant levels.

        If their capabilities are as limited as is indicated here, I fully expect they will meet those revenue estimates.

    1. Oliver – I don’t think United looks at the RM system purely as a cost center. Even for an airline as dysfunctional as United has been, that would be crazy. It’s possible they either didn’t know how to solve the problem or they just didn’t think the benefit could justify the cost. I’m not really sure, but it wouldn’t shock me if it was the former.

      1. United may outsource the care and feeding of the hardware, but even United owns and controls the secret sauce revenue management ‘Magic’, or lack there of.

        Only when the value add becomes a commodity (like load balancing systems from SABRE) are they a candidate to outsourcing.

        1. This is a CO system, not a UA system. This was a CO shortcoming.

          Yes, Yes, it’s UA now. Did pmUA have this issue with revenue?

  2. Airlines, for far too long, made it hard to buy anything but the cheapest fare on their websites and mobile apps.

    But this is basic forecasting, never use history as a guide if you know your history is tainted.

    Me thinks there must be something deeper that doesn’t make an investor day presentation.

  3. Reminds me of the heyday of Priceline. Nothing like bidding on what you were willing to pay and was truly amazing when scoring tickets for fractions of what the usual fares were. Not that I’d ever want to “bid” for a ticket on a business trip but yes, my price elasticity is much different for a business trip vs. leisure.

    So we know UA has a 20 year old system but is DL and AA doing it better?

    1. A – I think that’s important. While I don’t know how far along Delta and American are, this is certainly an issue they’ve had to struggle with as well. My guess is Kirby walked in and knew the problem right away, because he’s been trying to fix it in his previous life. I’d guess American is further along, but I don’t know for sure.

  4. @DavidM
    I’m sure there is value in noting that a ‘new sheriff in town’ can turn things around at UA but the detail that was presented shows that UA neglected investments in revenue mgmt. technology for years; that list wouldn’t exist and wouldn’t have the items on it that it does if they had invested in RM IT.
    It doesn’t take a Harvard mathematician to realize that the calendar is built on a 7 day week and demand changes every time an event moves relative to other days on the calendar. Ie… July 4 on a Thursday presents a different demand pattern than July 4 on a Monday. A revenue mgmt. system should be capable of storing enough history to reflect all of those permutations. Add in that events like Easter and Passover move by weeks every year and there has to be enough history to reflect all of that movement or the system can’t possibly accurately predict demand.
    Whether the RM system itself validates its own forecast or whether an external process is done to validate the system, it is inconceivable that an airline the size of UA did not recalibrate its own RM system’s forecast.
    Those are just 2 items on the list but those are major, major factors that should have never been allowed to remain unaddressed. There are processes to deal with every one of them and have been for years.
    UAL made the statement that their revenue performance is related to revenue mgmt. technology but not every airline has. I believe AA’s RM system is the same system it developed years ago but they apparently have done a better job of updating it; all technology has to be updated on a regular basis but they have acknowledged that their RM system is leaving some revenue on the table. Delta was the last of the legacy airlines to get serious about revenue mgmt. but has replaced its RM system several times and, I believe, is currently running an in-house designed version of Northwest’s system which has been tweaked several times since the merger; DAL’s revenue premium to the industry is directly tied to its ability to capture premium demand in the face of low cost competition. I believe LUV is replacing its RM system as part of its IT upgrade but they still have managed to recognize the limitations of the system they have.
    Revenue mgmt. technology especially for network based airlines (and LUV is just as much of a network airline as the big 3 – just without the multiple currencies that the big 3 operate in and must forecast for).
    The US airline industry and primarily American invented revenue mgmt. decades ago; there is no excuse why any of the big 3 shouldn’t have state of the art RM technology and it reflects poorly on decades of mgmt. to admit that their systems can’t incorporate some pretty basic principles of airline yield mgmt.

    1. I would give United enough credit to understand the rolling of the calendars for holidays and external factors such as major conferences, holidays etc. They aren’t that dumb.

      But buried in all the notes was that a high percentage of all flights have some element of revenue management done by hand. Humans. And many humans can’t Facebook and drive, never mind throw dozens of variables in their minds and optimize revenue. That is lso what United is looking to improve.

      1. I’m sure UA does realize the calendar issue but if the system doesn’t hold more than 2 years worth of history, the system can’t optimize based on calendar shifts; as you note, it has to be done manually which increases the risk of error, esp. if there is no forecast evaluation.
        It is very possible to evaluate analyst inputs to determine how accurate they are but if the system forecast isn’t validated, then it is certain they don’t validate analyst inputs.

        The size of the improvement which UA sees highlights precisely how big the problem is.

        and as noted, UA like AA and DL runs its own proprietary forecast engine; it is simply too risky to allow anyone else to know what is in the black box or have them be able to duplicate it which is what you get with commercially marketed revenue mgmt. systems. Thus, UA’s problem is not outsourcing of IT but a lack of investment in its own systems.

        There is abundant evidence to show that revenue mgmt. technology has about the fastest rate of return that any IT expenditure can generate.
        Nearly $1 billion in additional revenue attributable to IT improvements is enormous.

      2. Human skill and judgment are where/how you can theoretically help adjust and offset demand for differences in holiday timing etc.

  5. What might the implications of replacing Orion be for award availability? Will a new system enable more intelligent (stingy) opening of X, XN, etc.?

    1. Kyle – Award availability is more of a strategy set by the airline, so any change to that strategy about the importance of having seats open is going to have a greater impact than these particular tweaks.

        1. Kyle – Not knowing their current strategy, it’s hard to say. I don’t imagine that’s a big concern for the airline.

  6. Just because a lot of time and money is spent on revenue management, with a lot of attention by employees and consultants does not mean that they are getting it right. I think the tell has been that there have been so many last minute cheap tickets in economy and TOD and HOD upgrade buyups to business/first at OLCI with UA. I suspect that it is because their pricing is too high for too long far in advance, driving price-sensitive customers on competitive routes to OAL, and they overestimate how many remaining customers there are to buy higher fares or premium seats at the last minute. They are not very closely matching their supply for a seat at a particular price to the demand on a given day.

    But at least this indicates they are realizing that just because they have a system does not mean it is doing the right thing. How it turns out will be interesting to see.

    1. I believe you are exactly right. And it also raises the issue that revenue management systems or at least the demand data they produce should feed into the network planning systems. Humans can’t possibly calculate all of that accurately. Given UA’s propensity for many small RJS which don’t have favorable economics or capacity for low fare connections, their ability to match capacity with demand has been ?limited for years. They have a lot to fix but it will take time to put it all in balance

    2. Arthur – Actually, United’s belief is the opposite, that it isn’t holding back enough seats for higher fares.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Cranky Flier