Yves here. The individualized price gouging that Delta plans to implement seems to be coming soon. But as Hubert Horan explains, this case of “AI is coming to eat your lunch” does not look as easy to pull off as Delta might think.
Even so, Congresscritters are already saddling up to head Delta off at the pass. The proposed bill below is in addition to objections raised by three Senators that Hubert mentions below.
You shouldn’t be charged more or paid less because of your ZIP code or search history.
I just introduced a bill to ban “surveillance pricing” — when corporations use your data to hike up prices or cut your wages.
It’s invasive. It’s exploitative. And it needs to end. pic.twitter.com/4vOn6AFPF2
— Congressman Greg Casar (@RepCasar) July 27, 2025
The Economist just reported on another ruse that some US airlines tried implementing. It falls short of the individual pricing threat, but still got travelers’ dander up:
In recent years airlines have grown ever more sophisticated in their pricing techniques. American carriers’ latest method of singling out business passengers, though, is strikingly simple—and has sparked outrage in the travel blogosphere.
In May Thrifty Traveler, a website about travel bargains, reported that America’s three big legacy airlines—American, Delta and United—had started charging higher per-person fares for single-passenger bookings than for identical itineraries with two people. Kyle Potter, the author, grumbled that the practice amounted to carriers “weaponis[ing] their fares” against solo travellers who “can’t clone themselves”. Brian Kelly of The Points Guy, another travel site, called it “greed getting out of control” and said that airlines were “asking for government intervention”. Although no airline has yet commented on the subject, Delta and United reportedly scrapped the practice amid the criticism.
To investigate further, The Economist turned to Serpapi, an automated interface to Google Flights, a fare database. For all direct domestic journeys on America’s three big legacy carriers, we downloaded one-way main-economy fares as of July 20th for one and two passengers to travel on Monday July 28th, choosing whichever flight had an airline’s cheapest single-passenger price for that route that day. We also pulled return fares—one the following Friday and the other on Saturday—for both one and two passengers. In total, we amassed 19,000 prices across 3,200 routes.
Delta has indeed abandoned the technique: its two-passenger price is always at least twice the fare for one. But American and United have persisted. Solo flyers who travel with them within the work week can end up paying more than everyone else, including solo travellers whose journey involves a weekend stay and those who travel with others, regardless of whether their trip stretches into a weekend.
In other words, airlines are exposed to consumer backlash, and even, as regulated entities, political pressure.
So individualized pricing were to get traction, some airlines could tout their abstention. I could see Middle Eastern carriers, like Qatar Air and Emirates, which cater to a high end clientele, advertising that they don’t nickel and dime customers this way.
By Hubert Horan, who has 40 years of experience in the management and regulation of transportation companies (primarily airlines) and has been publishing analysis of Uber since 2016
Delta’s President Glen Hauenstein told investors that it sees AI-driven personalized (or surveillance) pricing as “a full re-engineering of how we price, and how we will be pricing in the future”. He said “we’re really excited about partnering with Fetchrr” (Delta’s Israel-based AI consultant) whose tools were helping Delta “get inside the mind of our consumer” so that eventually, “we will have a price that’s available on that flight, on that time, to you, the individual.” He said that AI tools were now setting prices on 3% of Delta flights but he expected 20% of flights would be AI priced by the end of 2025.[1]
The definition of personalized/surveillance pricing is well understood and represents a radical departure from other forms of algorithmic pricing. [2] Algorithmic pricing simply means that many of the rules (algorithms) governing when prices should be adjusted in response to new information about consumer demand or competitor actions have been automated. Airlines pioneered pricing automation in the 1980s, which incorporated second degree price discrimination, so different fares can be offered on flights with higher or lower demand, and different fares can be offered with different conditions (advance purchase requirements, different checked baggage or seat assignment surcharges).
But under second degree price discrimination every consumer sees the exact same price offering at any given time. Every consumer has the identical opportunity to pay less if they buy in advance and can only sit in the very back of the plane or pay more at the last minute and sit closer to the front. Delta is reengineering its pricing around using first degree price discrimination. Personalized/surveillance data is used to estimate the maximum price each individual would be willing to pay. Individuals would no longer see the same prices that all others are seeing.
Any claims that first degree price discrimination would improve overall efficiency or benefit consumers are false. The original use of second degree price discrimination in the 80s did improve efficiency and consumer welfare. Price sensitive customers who were willing to buy tickets on lower demand in advance got much lower fares than previously available, and more time-sensitive customers willing to pay more got greater access to higher demand flights at the last minute. Airlines achieved much higher capacity utilization (load factors), earned much more revenue on flights that had previously been half empty, and avoided the huge expense of adding aircraft that were only needed at peak periods.
But the efficiency gains from price discrimination were exhausted long ago. Increased price discrimination would only be a wealth transfer from consumers to airline shareholders. To use simplistic Economics 101 terminology, when everyone sees a standard market price, a consumer surplus is created because some consumers would have been willing to pay more. First degree price discrimination using personalized/surveillance pricing attempts to eliminate as much of that consumer surplus as possible by charging each customer the most they would be willing to pay. Passengers pay more without getting anything of value in return.
Lots of companies have tried to introduce personalized/surveillance pricing but corporate announcements including Delta’s usually generate consumer backlash. Three Democratic Senators complained that this could support predatory pricing and demanded that Delta disclose what information about individual consumers it would be using to set prices. [3] Airline passengers understand that airlines don’t have their best interests in mind, can quickly see through attempts to claim first degree price discrimination is the same thing as traditional second-degree discrimination and readily dismiss ludicrous attempts to claim personalized/surveillance pricing will actually lead to lower prices. [4]
The shift from second degree to AI-driven first-degree price discrimination would be one of the biggest aviation changes in decades. So I went to the Fetcherr website and found surprisingly that it said that its airline AI offerings don’t actually do any of the things Hauenstein was claiming. [5] They don’t collect data about individual customers or analyze the price elasticities or other characteristics of individual customers or fine-tuned customer segments. Fletchrr emphasizes that its AI/data tools are designed to be easily integrated into existing pricing practices and there is absolutely no suggestion they were designed to help an airline to completely reengineer its pricing function.
Something is seriously wrong here, but what? Are Delta and/or Fletchrr being less than honest about what they are doing?
Major investments in AI tools are only appropriate in cases where a company needs to process massively greater amounts of external data than before and wants to divine correlations that human analysts using PC based software could never find. Large Language Models identify text patterns that can help automate preparation of article summaries and documents. Activity specific models find correlations between millions of historical inputs and outputs that can support the automation of basic tasks like coding or accounting.
Two major AI applications are especially relevant here. The primary “get inside the mind of the consumer” case is advertising, where large companies like Google and Facebook used their access to massive amounts of data about individual users to program real-time ads more effectively than traditional human marketing analysts. The primary “use Big Data to totally re-engineer previously human+PC software analytical functions” came in the 1990s when hedge funds used hitherto unprecedented amounts of computing power to find factors correlated with asset price changes that drove more higher average returns and incorporate them into high-speed computerized trading strategies. [6]
One possible explanation for these apparent contradictions is that Fletchrr doesn’t understand airline pricing very well. Its founders come from the hedge fund world where AI tools are critical because there are millions of traders, billions of transactions, individual markets are highly volatile, can be influenced by an unpredictable range of external factors, and major traders are constantly changing their strategies. Fletchrr never explains why this experience would apply to airlines where supply and demand are highly stable in the short/medium term, the number of airline competitors is limited and while their pricing approaches can evolve over time they haven’t changed dramatically or unexpectedly in decades.
Why did Fletchrr see airlines as a prime target for its AI sales efforts? Because they thought airlines were “outdated” “undisrupted” and had seen few recent technological advances.
In fact, airlines were perhaps the non-military industry quickest to drive innovation in operations research, IT technology, and modern pricing/distribution tools, and of course saw business models totally transformed in the last quarter of the 20th Century. They may not seem highly dynamic to someone whose experience is limited to the recent development of hedge fund quant models, but very few industries anywhere experience that rate of technological change.
Fletchrr says airlines need AI tools to handle today’s faster rate of change, but the only examples of hard to handle change it offered were nonsensical–the impact of Covid on demand and the challenge of previously unscheduled flights into Doha for the FIFA World Cup. Fletchrr doesn’t provide any concrete examples of major problems it can solve that current pricing systems can’t deal with. The input data for Fletchrr’s AI market model is the exact same input data airline revenue management systems have been using for decades.
There might be value in a tool that can quickly process greater volumes of input data; every corporate function could be improved at the margin, and perhaps these gains would justify the IT investment. But these would be marginal improvements, would not represent the complete reengineering of pricing that Delta claimed, or anything that could produce the major revenue/profit impacts that would justify major announcements to investors.
Another possible explanation is that Delta executives never thought through the requirements, potential gains and implementation risks of re-engineering its existing systems into personalized/surveillancepricing. Perhaps Delta never bothered to figure out that Fletchrr’s software was only offering marginally greater automation of traditional pricing tasks and never considered that Fletchrr’s software wasn’t designed to drive the dramatic changes it thought its investors would value. Delta hasn’t made any attempt to define the shortcomings in its existing pricing systems that it is hoping to address or explain how a future reengineered system would differ from today’s.
Maybe Delta executives had drunk the Kool-aid of the AI hype machine, assumed anything labelled “AI” would have magical, powerful impacts anywhere and never stopped to consider whether the conditions that allowed AI tools to create value in other industries applied here. Perhaps Delta executives more cynically thought that splashy announcements of big AI projects would juice the stock price, would reinforce Delta’s image of having more progressive management than United or Delta, and assumed that investors would never hold management accountable if there wasn’t a big revenue/profit boost.
Even if it is I possible that Fletchrr’s hedge fund trained AI experts overestimated the applicability of its tools to airlines and that Delta management accepted too much AI hype it seems rather improbable that both these two sophisticated companies would announce a major effort where the goals were badly misaligned.
One more plausible explanation is that Fletchrr fully understands that Delta is determined to achieve first degree price discrimination, and both parties wanted to obscure this. It could be that Fletchrr knew its AI-driven Market Model was well suited for personalized/surveillance pricing but deliberately excluded any mention of this from its promotional material to help shield airline clients from external criticism in cases like this. Fletchrr’s website gives Delta a way to plausibly deny outside critics (“that’s not what these AI tools do!”) without having to formally disavow their pursuit of personalized/surveillance pricing. It could be that Fletchrr’s models were never designed to process huge volumes of personal information but Delta believe they can be adopted to support first degree price discrimination.
It remains possible that Delta’s effort to use AI-driven personalized/surveillance pricing will fail to materially boost profitability. The articles reporting Hauenstein’s claim mentioned a range of potential obstacles to personalized/surveillance pricing including the ready availability of data on market (non-personalized) prices that would allow flyers to see if Delta was trying to get them to pay above-market fares.
While it is widely understood how Google and Facebook can use terabytes of personal data to tailor ad displays, no one has publicly explained how personal data would allow an airline to calculate price elasticities for each customer and reliably predict that this individual shopping for this specific flight would be willing to pay more than it was asking other customers to pay.
It is not even clear that the price elasticities of individual customers can be measured, or that a system could identify how an individual’s elasticity varied from trip to trip (e.g. critical last minute sales meeting, attending a conference that may or may not have value, taking the kids to visit grandma).
Many observers assume that first degree price discrimination would require forcing most passengers to not only grant Delta access to much more personal information than they have now, but to force them use Delta-controlled sales channels. A recent American attempt to force corporate agents serving higher yielding passengers to use a captive channel actually reduced revenue by over a billion dollars and was withdrawn. [7] Delta’s best customers might similarly resist any attempt to force them to use a channel that prevented them from seeing what true market rates were.
Uber provides a case example of a company where shifting to first degree price discrimination did produce a very large profit boost. Uber previously offered the same fare to any customer and offered the same payment to any driver (based on factors such as distance and time of day) with a system that estimated the highest fare/lowest payments they would accept.[8]
But if Delta was attracted to personalized/surveillance pricing by the big profit boost Uber achieved it may be badly disappointed because Delta has none of the structural advantages that allow Uber to maximize exploitive discrimination. Uber rides are last minute purchases and riders have no ability to compare prices. There are no independent Google/Kayak/Expedia-type sources of true market taxi pricing information. Delta frequent flyers understand airline pricing and would quickly figure out if changes were unfavorable. Uber users have no real idea how Uber pricing works, and Uber (and Lyft) have achieved quasi-monopoly pricing power after using billions in predatory subsidies to drive independent competition out of the market.
It should be emphasized that the central issue, should Delta have any success pursuing first degree price discrimination is not “AI technology” or “pricing algorithms” but the ability to exploit anti-competitive market power.
Simplistic Economics 101 models say producers can’t achieve the capture of consumer surplus that Delta hopes to achieve using first degree price discrimination because consumers would be protected by market competition and the availability of perfect information about market prices. First degree price competition works for Uber because they tightly control all marketplace information, have eliminated all meaningful competition and any ability of elected officials to enforce consumer and labor law protections. Even Uber investors have no ability to see how pricing and driver compensation changes affect profitability.
Thus any attempt to implement first degree price discrimination requires subverting the proper workings of competitive markets.
Without significant artificial market power Delta would have no ability to force its best customers to use Delta controlled distribution channels, or to limit their ability to see if Delta is only showing them fares higher than other customers can get. Without significant artificial market power Delta would have no ability to blow off customer backlash, negative publicity and complaints from Congress. The ability to impose first degree price discrimination should perhaps be seen as prima facie evidence that a company has artificial anti-competitive market power.
While normal companies that achieved a multi-billion profit recovery would aggressively publicize the brilliance of its management moves, Uber management has gone to great lengths to keep the public from becoming aware that it made a major shift to first order price discrimination and that that shift was the major reason Uber finally became profitable. Uber understands that a greater awareness of passenger/driver exploitation could not only drive serious customer/political backlash but create awareness that its profitability had nothing to do with management brilliance but was driven entirely by the subversion of market competition.
Delta might rationally believe that it already has the artificial market power to pursue personalized/surveillance pricing while ignoring the objections of its best customers but Uber management might warn them to take these risks more seriously. American’s CEO has already called Delta’s proposed pricing shift a “bait and switch” plan. [9]
_______
[1] Hauenstein’s original comments about AI pricing were made at a Delta “Investor Day” last November, but there was no press coverage until he made follow up comments during a recent quarterly earnings call. Gary Leff, Delta Is Turning Ticket Pricing Over To AI—By Year-End, 20% Of Fares Will Exactly Match The Most You’re Willing To Spend, View From The Wing, 10 July 2025; Irina Ivanova, Delta moves toward eliminating set prices in favor of AI that determines how much you personally will pay for a ticket, Fortune 16 July 2025; Matt Novak, Delta Set to Expand AI-Powered Dynamic Ticket Pricing by the End of 2025, Gizmodo, 17 July 2025;
[2] For a review of over 300 journal articles discussing both traditional algorithmic and personalized pricing see Seele, P., Dierksmeier, C., Hofstetter, R. et al. Mapping the Ethicality of Algorithmic Pricing: A Review of Dynamic and Personalized Pricing. J Bus Ethics 170, 697–719 (2021).
[3] Michael Kan, Wendy’s Clarifies Digital Surge-Pricing Strategy After Blowback, PCMag, February 28, 2024; Cory Doctorow, Surveillance pricing lets corporations decide what your dollar is worth, Pluralistic 24 Jun 2025; David Shepardson, Delta plans to use AI in ticket pricing draws fire from US lawmakers, Reuters, July 23, 2025.
[4] Gary Leff, Airlines Now Quietly Let AI Set Ticket Prices—Surprisingly, That’s Great News For Your Wallet, View From The Wing, 15 July 2025
[5] https://www.fetcherr.io/technology which includes a white paper about its Large Market Model a 45 minute talk about Fetcher’s AI tools apply to airlines by AI Chief Uri Yerushalmi, and a You Tube video interview with Virgin Atlantic VP of Pricing and Revenue Management Chris Wilkinson about his use of Fletchrr’s tools. Fletchrr says Azul, WestJet, Virgin Atlantic, and VivaAerobus are also clients of its AI airline pricing tools.
[6] For the story of Jim Simons, Robert Mercer, Renaissance Technologies and the development of algorithmic trading by hedge funds see Gregory Zuckerman, The Man Who Solved The Market: How Jin Simons Launched the Quant Revolution, New York, Penguin Books, 2019
[7] Benjamin Zhang, American Airlines CEO Admits It Messed up Ticket-Sales Strategy Change, Business Insider, May 29, 2024, Justin Dawes, American Airlines Recovering After Failed Direct Bookings Strategy, November 12, 2024
[8] Here are a couple recent articles about external studies of Uber’s radical new algorithmic approaches: Simon Goodley, Rough ride: how Uber quietly took more of your fare with its algorithm change, The Guardian 19 June 2025; Simon Goodley, Second study finds Uber used opaque algorithm to dramatically boost profits, The Guardian 25 June 2025. For a fuller explanation of Uber’s financial turnaround see Hubert Horan: Can Uber Ever Deliver? Part Thirty-Five: What Drove Uber’s Recent $8 Billion P&L Improvement?, Naked Capitalism, 25 Feb 2025
[9] Christine Boynton, American Airlines CEO Blasts ‘Other’ AI Talk: ‘This Is Not About Tricking’, Aviation Week, 24 July 2025
