Understanding Algorithmic Pricing

You decide to buy the plane ticket only to get an email a few days later from the airline's rewards program letting you know an upcoming sale will cut the price for your chosen route by $100. Or you hold off, hoping for a sale, only to check a few days later and see the price has increased by $100. In both scenarios, a deep feeling of regret sinks in. But in an equally plausible scenario, you held off, and the airline sent you a message: "We noticed you were shopping for this flight but didn't book it. It's now on sale!" You book right away with a smile on your face and some extra money in your pocket.

For everyday consumers, buying plane tickets can be a dizzying experience. Prices are always going up and down, seemingly changing every day, if not more frequently. Nearly every airline has a rewards program, and deals are perpetually coming and going. And it's not just airlines. So many goods, from concert tickets and gasoline to groceries and home furnishings, all seem to have exclusive sales, membership discounts, hidden fees, and changing prices driven by a wide range of factors that are nearly impossible to understand. It's a far cry from the days when it seemed the prices for most goods were universally applied, with well-publicized sales and coupons available to everyone.

In recent years, technology has facilitated the more rapid and individualized adjustment of prices. Whereas before this might have required an army of pricing specialists working around the clock, businesses can now use an algorithm (a set of instructions) to automatically set and reset prices through the ongoing incorporation of a wide range of inputs. Those inputs can range from basic supply and demand levels, competitor prices, weather, and the behavior and characteristics of individual consumers.

Proponents of algorithmic pricing argue that it increases market efficiency, allowing better real-time matching of prices to goods, which can benefit both consumers and retailers. They also suggest the technology will simplify vendors' pricing decisions and increase competition, leading to lower prices as firms' awareness of real-time competitor pricing is likely to increase.

Other observers, however, maintain that it is fraught with questions of fairness. It is illegal, for example, to discriminate based on an individual's race or gender, but algorithms can set prices for an individual based on other data, such as income, residence, and age. Firms can use that data to potentially determine the highest price each customer might be willing to pay for a product or a service. Recent economic research also suggests that algorithmic pricing creates opportunities for competing firms to collude, even if unintentionally, resulting in bigger profits for firms and higher prices for consumers.

What Is Algorithmic Pricing and Who Is Using It?

When a firm uses algorithmic pricing, inputs like those mentioned above are loaded into computer programs of varying complexity with instructions about how each data point should be incorporated, and then the algorithm generates product prices across time, location, or even consumers.

These inputs already informed firms' pricing decisions to varying degrees prior to the advent of the technology, but the algorithms allow those pricing decisions to happen automatically, lowering the costs associated with changing prices, which saves firms time and money. Advanced artificial intelligence (AI) models can use machine learning, predictive analytics, or large language models to process huge amounts of data and make pricing decisions in real time. Of course, the technology isn't free, but a costlier, more advanced algorithm can allow retailers to gain an advantage over their competition.

These inputs already informed firms' pricing decisions to varying degrees prior to the advent of the technology, but the algorithms allow those pricing decisions to happen automatically, lowering the costs associated with changing prices, which saves firms time and money. Advanced artificial intelligence (AI) models can use machine learning, predictive analytics, or large language models to process huge amounts of data and make pricing decisions in real time. Of course, the technology isn't free, but a costlier, more advanced algorithm can allow retailers to gain an advantage over their competition.

"It's a competitive advantage for each retailer. They're looking at what their customers are telling them about what they want and need. They're looking at supply chains and global disruptions," says Jason Straczewski, group vice president of government relations and political affairs at the National Retail Federation. "They're looking at what their competitors are doing in the marketplace and what they're charging. They're asking, 'How can I undercut them and be the most competitive price on the street?' So much goes into it."

There are two general categories of algorithmic pricing. First, dynamic pricing entails automated price changes resulting from broad economic inputs such as market demand, inventory levels, and competition. Alternatively, personalized pricing (also known as surveillance pricing) involves charging individuals different prices based on their personal behavior or attributes, which are usually gathered from a company's rewards program or by tracking online behavior and spending patterns. This method often relies more on AI than basic dynamic pricing, as AI can process more data at greater frequencies and speeds.

Large financial institutions are also using it for a range of pricing activities. For example, JPMorgan Chase first used AI pricing technology in its consumer banking activity, incorporating personal characteristics and other information like creditworthiness, risk profiles, and market conditions when setting individual mortgage and loan rates. It has since expanded its algorithmic pricing to other services it offers, including wealth management, investment products, and standard banking fees.

Jonathan Adams is a senior economist at the Kansas City Fed, and he is one of the authors of the Journal of Monetary Economics article. He notes that firms are also beginning to use algorithmic pricing for business-to-business transactions, and that many of those transactions include the selling of pricing algorithms themselves. "A lot of what we see in the adoption of AI pricing is actually selling AI pricing as a service," he says. "Some of the top firms that are doing this are consultancies that are trying to find people who can work on these algorithms and then provide them as a service to their clients."

What Happens When There Is No Uniform Price?

Algorithmic pricing can facilitate greater efficiency in the market. A 2025 policy brief by the Mercatus Center, a free-market-oriented think tank affiliated with George Mason University, noted that dynamic pricing does so by "incentivizing consumption during low-demand periods and disincentivizing it when there is an excess of demand relative to supply." In other words, it can allow for more targeted reduction in prices during periods of sluggish demand or raise prices when demand is high.

Dynamic pricing can also benefit consumers by reserving supply for those who need or value it the most. In a 2022 Econometrica article, Yale University economist Kevin Williams demonstrated that for airline markets, dynamic pricing, relative to a single, uniform price, benefits leisure travelers who purchase tickets early at the expense of business travelers who frequently book at the last minute. Vacationers tend to look for the best deal and are more flexible about when they travel, while business travelers need to travel on certain days and often book close to their departure date. But most importantly, they are willing to pay a premium for those last-minute flights, which allows the airline to offer lower prices to the early booking leisure travelers. Williams found that overall welfare is greater under this dynamic pricing system than uniform pricing, with algorithms allowing for more frequent adjustments that maximize utility for both early purchasing consumers and firms.

Perhaps the key change stemming from algorithmic pricing shows up in personalized pricing. Before the wide adoption of algorithmic pricing, retailers began using price tags in the 19th century, fixing one price to an item for all customers. Although retailers often held sales on particular days and at specific times to attract certain demographics who were more likely than others to be available, their ability to carefully target pricing was relatively limited.

By collecting data on consumers' behavior, spending habits, and locations, algorithms can estimate how much each customer might be willing to pay for a particular good or service. Imagine, for example, that an individual visits a retailer's website for the first time and purchases a household item like a vacuum cleaner at the listed price. That consumer did not receive a discount but needed the vacuum quickly and is satisfied with the price. He or she maybe would have even been willing to pay more.

Now, imagine another individual who regularly shops at that same store, belongs to its rewards program, and is much more budget-conscious, often looking around for the best deals. That customer visits the website at the same time as the other customer, sees the price, and wonders if there's a better deal elsewhere. As he or she looks at the prices on other websites or online stores, the original store detects they are shopping around and notifies the customer that the price has dropped. So, the second customer buys the same vacuum from the same store but at a lower price.

"That's the competitive environment American retailers are up against. There's a lot of choice," says Straczewski. "The customer has nearly limitless options to find what they're looking for, so retailers are using this data to serve you in a somewhat personalized environment because they want your loyalty and trust. They want you to keep shopping with them and not their competitors. That's how they're leveraging this data."

Both customers and the retailer are satisfied with this scenario. When there was a single uniform price, which could have been below what some customers were willing to pay, those consumers effectively pocketed that difference — a value economists call "consumer surplus." In the world of personalized pricing where a firm can charge a higher price to those with a higher willingness to pay, that surplus transfers to the firm, and that, in turn, allows the firm room to lower the price for the more price-sensitive consumers and still increase total profit.

Fairness Concerns

In the cases of dynamic and personalized pricing, price discrimination allows those who are more price-sensitive to seek out, and perhaps find, lower prices. However, it is also true that some consumers end up paying more for a good or service than others.

This raises questions of fairness in the minds of many consumers when it comes to vendors' use of pricing algorithms. In an International Journal of Research in Marketing article, researchers from over a dozen universities around the world explored those perceptions in detail. They noted three specific concerns that consumers might have over the rules firms use when developing their algorithms. First, compared to a human price-setter, people may perceive more frequent price changes as unfair. Second, while consumers are aware that a product's price may change in response to the seller's costs, they may not be aware of how other inputs such as inventory levels, supply and demand shifts, and other factors might lead to price increases (or price changes more broadly). Finally, consumers' comfort with algorithmic pricing can be context-dependent: They may be used to and accept dynamic pricing when it comes to travel or entertainment services after seeing it for years, but the technology may not yet be normalized when shopping for household goods or banking.

These concerns are heightened when consumers believe or see that other people are paying different prices for the same goods, which can also appear unfair, as well as when they see the price rise or fall after they have already made their purchase. This perception of unfairness can persist even if those price differences result from the price discrimination logic and personalized pricing described above. 

Policymakers are responding to this unease. As of this writing, New York and Connecticut have enacted laws requiring businesses to notify consumers whenever a price is determined by an algorithm using personal data, and at least 72 bills across 31 states, Washington, D.C., and Congress have been introduced to mandate disclosure, prohibit outright the use of personalized pricing, or study the practice. At the federal level, the Federal Trade Commission (FTC) is developing plans to limit personalized pricing, as it recently released a request for public comment for a proposed policy change whereby personalized pricing would be in violation of the FTC Act if a seller misrepresents a price as uniform when it is actually personalized. The FTC also wants to require companies to disclose when they are using personal data and to get consumers' consent when they collect it.

The Potential for Collusion

Beyond concerns over consumers being squeezed and the collection of personal data, research has found that rather than leading to lower prices, algorithmic pricing can facilitate higher prices when firms are competing for buyers.

Historically, firms have been able to collude on price-setting, even absent (illegal) explicit agreements, without the presence of algorithmic pricing technology. They can, for example, converge on a price set by a dominant firm. This practice is common in the airline and gasoline industries, and if other firms match that price, they can all be rewarded with higher prices and bigger profits. However, if in this scenario the responding firms decide to defect from the tacit coordination and lower prices, undercutting the firm that first raised prices, it can lower prices in return, leading to a price war. This incentivizes firms to adhere to a "match price increases, punish price cuts" ethos that keeps prices higher.

This same outcome can happen when firms use pricing algorithms. In a 2020 American Economic Review article, economists at Luiss Guido Carli University, the European University Institute, and the University of Bologna ran a simulated pricing game and found that the algorithms learned on their own without any instruction that a "reward or punish" strategy allowed them to maintain supracompetitive prices, or those that otherwise would not exist in a truly competitive market.

While this finding took place in a repeated game simulation, in a 2023 American Economic Journal: Microeconomics article and a current working paper, economists Zach Brown of the University of Michigan and Alexander MacKay of the University of Virginia showed how prices among competitors can remain elevated by giving a single competitor's algorithm instructions to react quickly to a rival's price.

"You can set your algorithm so that, if your rival doesn't choose a high enough price, you immediately undercut or punish them," says MacKay. "That may sound like intense competition, but the result is that the rival stops choosing low prices because they know that your algorithm is going to respond with prices that are even lower. The threat of this 'price-war-like' outcome keeps prices higher for everyone. Likewise, if the algorithm matches price increases, this can give the rival an incentive to raise prices."

Empirical research has shown that algorithmic pricing can induce price increases by firms in competitive environments. In a 2024 Journal of Political Economy article, researchers in Canada, Britain, and Spain looked at the German retail gasoline market and found that after algorithmic pricing software was first introduced in 2017, mean station-level prices and margins increased by about 1.3 cents per liter, or a marginal increase of about 15 percent. In areas where a gas station using the technology had no competition, mean prices and margins saw no statistically significant change, but in competitive markets (two or three stations), margins increased by 3.2 cents per liter, or about 38 percent. These researchers found that these increases appeared about one year after adoption, suggesting the algorithms learned that dropping prices or attempting to undercut the competition was not a long-run profitable strategy.

MacKay stresses that these outcomes hinge on a firm's algorithm being given instructions to incorporate competitors' prices into its decisions. It differs from the traditional legal sense of collusion, where competitors agree to set prices at or above a certain point. Instead, firms instruct their algorithm to unilaterally react to rivals' prices. Brown and MacKay call the extreme form of this behavior "coercion." To stop algorithms from raising prices, policymakers and regulators may need new solutions. MacKay notes, "It raises the question: Should algorithms be able to use rivals' prices at all?"

Given that competitors' prices are a key input in firms' pricing decisions and the difficulties that would come with policing firms' algorithms, restricting the use of that information would likely face strong opposition. That leaves consumers and firms with more work to do to build trust in a rapidly evolving pricing environment.

Readings

Brown, Zach, and Alexander MacKay. "Competition in Pricing Algorithms," American Economic Journal: Microeconomics, May 2023, vol.15, no. 2, pp. 109-156.

Calvano, Emilio, Giacomo Calzolari, Vincenzo Denicolò, and Sergio Pastorello. "Artificial Intelligence, Algorithmic Pricing, and Collusion," American Economic Review, Oct. 2020, vol. 110, no. 10, pp. 3267-3297.

Taylor, Cody. "The Case for Algorithmic Pricing: Consumer Welfare, Market Efficiency, and Policy Missteps." The Mercatus Center at George Mason University, May 14, 2025.

Williams, Kevin. "The Welfare Effects of Dynamic Pricing: Evidence from Airline Markets," Econometrica, March 2022, vol. 90, no. 2, pp. 831-858.