Why Do Gas Station Prices Constantly Change? Blame the Algorithm
Retailers are using artificial-intelligence software to set optimal prices, testing textbook theories of competition; antitrust officials worry such systems raise prices for consumers
The Knaap Tankstation gas station in Rotterdam, Netherlands, uses a2i Systems artificial-intelligence pricing software. Photo: Knaap Tankstation BV
Updated May 8, 2017 6:41 p.m. ET
ROTTERDAM, the Netherlands—One recent afternoon at a Shell-branded station on the outskirts of this Dutch city, the price of a gallon of unleaded gas started ticking higher, rising more than 3½ cents by closing time. A little later, a competing station 3 miles down the road raised its price about the same amount.
The two stations are among thousands of companies that use artificial-intelligence software to set prices. In doing so, they are testing a fundamental precept of the market economy.
In economics textbooks, open competition between companies selling a similar product, like gasoline, tends to push prices lower. These kinds of algorithms determine the optimal price sometimes dozens of times a day. As they get better at predicting what competitors are charging and what consumers are willing to pay, there are signs they sometimes end up boosting prices together.
Advances in A.I. are allowing retail and wholesale firms to move beyond “dynamic pricing” software, which has for years helped set prices for fast-moving goods, like airline tickets or flat-screen televisions. Older pricing software often used simple rules, such as always keeping prices lower than a competitor.
On the Same Track
Two competing gas stations in the Rotterdam area both using a2i Systems pricing software roughly mirrored each other’s price moves during a selected week.
Hourly change in price for unleaded Euro 95 gasoline;
three-hour moving average for hours that both stations were open*
1.5 U.S. cents a gallon
These new systems crunch mountains of historical and real-time data to predict how customers and competitors will react to any price change under different scenarios, giving them an almost superhuman insight into market dynamics. Programmed to meet a certain goal—such as boosting sales—the algorithms constantly update tactics after learning from experience.
Ulrik Blichfeldt, chief executive of Denmark-based a2i Systems A/S, whose technology powers the Rotterdam gas stations, said his software is focused primarily on modeling consumer behavior and leads to benefits for consumers as well as gas stations. The software learns when raising prices drives away customers and when it doesn’t, leading to lower prices at times when price-sensitive customers are likely to drive by, he said.
“This is not a matter of stealing more money from your customer. It’s about making margin on people who don’t care, and giving away margin to people who do care,” he said.
Driving the popularity of A.I. pricing is the pain rippling through most retail industries, long a low-margin business that’s now suffering from increased competition from online competitors.
“The problem we’re solving is that retailers are going through a bloodbath,” said Guru Hariharan, chief executive of Mountain View, Calif.-based Boomerang Commerce Inc., whose A.I.-enabled software is used by StaplesInc. and other companies.
Staples uses A.I.-enabled software to change prices on 30,000 products a day on its website. Photo: Richard B. Levine/ZUMA PRESS
The rise of A.I. pricing poses a challenge to antitrust law. Authorities in the EU and U.S. haven’t opened probes or accused retailers of impropriety for using A.I. to set prices. Antitrust experts say it could be difficult to prove illegal intent as is often required in collusion cases; so far, algorithmic-pricing prosecutions have involved allegations of humans explicitly designing machines to manipulate markets.
Officials say they are looking at whether they need new rules. The Organization for Economic Cooperation and Development said it plans to discuss in June at a round table how such software could make collusion easier “without any formal agreement or human interaction.”
“If professional poker players are having difficulty playing against an algorithm, imagine the difficulty a consumer might have,” said Maurice Stucke, a former antitrust attorney for the U.S. Justice Department and now a law professor at the University of Tennessee, who has written about the competition issues posed by A.I. “In all likelihood, consumers are going to end up paying a higher price.”
In one example of what can happen when prices are widely known, Germany required all gas stations to provide live fuel prices that it shared with consumer price-comparison apps. The effort appears to have boosted prices between 1.2 to 3.3 euro cents per liter, or about 5 to 13 U.S. cents per gallon, according to a discussion paper published in 2016 by the Düsseldorf Institute for Competition Economics.
Makers and users of A.I. pricing said humans remain in control and that retailers’ strategic goals vary widely, which should promote competition and lower prices.
“If you completely let the software rule, then I could see [collusion] happening,” said Faisal Masud, chief technology officer for Staples, which uses A.I.-enabled software to change prices on 30,000 products a day on its website. “But let’s be clear, whatever tools we use, the business logic remains human.”
Online retailers in the U.S., such as Amazon.com Inc. and its third-party sellers, were among the first to adopt dynamic pricing. Amazon.com declined to comment.
Since then, sectors with fast-moving goods, frequent price changes and thin margins—such as the grocery, electronics and gasoline markets—have been the quickest to adopt the latest algorithmic pricing, because they are the most keen for extra pennies of margin, analysts and executives say.
The pricing-software industry has grown in tandem with the amount of data available to—and generated by—retailers. Stores keep information on transactions, as well as information about store traffic, product location and buyer demographics. They also can buy access to databases that monitor competitors’ product assortments, availability and prices—both on the web and in stores.
A.I. is used to make sense of all that information. International Business Machines Corp. said its price-optimization business uses capabilities from its Watson cognitive-computing engine to advise retailers on pricing. Germany’s Blue Yonder GmbH, a price-optimization outfit that serves clients in the grocery, electronics and fashion industries, said it uses neural networks based on those its physicist founder built to analyze data from a particle collider.
Neural networks are a type of A.I. computer system inspired by the interconnected structure of the human brain. They are good at matching new information to old patterns in vast databases, which allows them to use real-time signals such as purchases to predict from experience how consumers and competitors will behave.
Algorithms can also figure out what products are usually purchased together, allowing them to optimize the price of a whole shopping cart. If customers tend to be sensitive to milk prices, but less so to cereal prices, the software might beat a competitor’s price on milk, and make up margin on cereal.
“They’re getting really smart,” said Nik Subramanian, chief technology officer of Brussels-based Kantify, who said its pricing software has figured out how to raise prices after it sees on a competitor’s website that it has run out of a certain product.
Algorithmic pricing works well in the retail gasoline market, because it is a high-volume commodity that is relatively uniform, leading station owners in competitive markets to squeeze every penny.
For years, price wars in cutthroat markets have followed a typical pattern. A retailer would cut prices to lure customers, then competitors would follow suit, each cutting a little more than the others, eventually pushing prices down close to the wholesale cost. Finally one seller would reach a breaking point and raise prices. Everyone would follow, and the cycle started all over.
Some economists say the price wars helped consumers with overall lower prices, but led to very thin margins for station owners.
Danish oil and energy company OK hired a2i Systems in 2011 because its network of gas stations was suffering from a decade-old price war. It changed what it charged as many as 10 times a day, enlisting a team of people to drive around the country and call in competitors’ prices, said Gert Johansen, the company’s head of concept development.
A2i Systems—the name means applied artificial intelligence—was started by Alireza Derakhshan and Frodi Hammer, both engineering graduates of the University of Southern Denmark, in Odense. Before focusing on fuel, they built other A.I. systems, including a game displayed on interactive playground floor tiles that adapted to the speed and skill level of the children running around on top.
For OK, a2i created thousands of neural networks—one for each fuel at each station—and trained them to compare live sales data to years of historical company data to predict how customers would react to price changes. Then it ran those predictions through algorithms built to pick the optimal prices and learn from their mistakes.
In a pilot study, OK split 30 stations into two sets, a control group and an a2i group. The group using the software averaged 5% higher margins, according to a paper Mr. Derakhshan presented last June at an A.I. conference in Seville, Spain. (scroll down to continue reading…)
Scandinavian supermarket chain REMA 1000 says it will roll out price-optimization software made by Texas-based Revionics Inc. in coming months. Photo: Joseph Dean/Newscom/ZUMA Press
The new system could make complex decisions that weren’t simply based on a competitor’s prices, Mr. Derakhshan said in an interview.
One client called to complain the software was malfunctioning. A competitor across the street had slashed prices in a promotion, but the algorithm responded by raising prices. There wasn’t a bug. Instead, the software was monitoring the real-time data and saw an influx of customers, presumably because of the long wait across the street.
“It could tell that no matter how it increased prices, people kept coming in,” said Mr. Derakhshan.
On the outskirts of Rotterdam, Koen van der Knaap began running the system on his family-owned Shell station in recent months. Down the road, a station owned by Tamoil, a gasoline retailer owned by Libya’s Oilinvest Group, uses it too.
During a late-March week for which both Tamoil and Mr. van der Knaap provided hourly data, the costs for unleaded gas at the two stations—which vary in opening hours and services—bounced around independently much of the time, and generally declined, reflecting falling oil prices that week.
During some periods, however, the stations’ price changes paralleled each other, going up or down by more than 2 U.S. cents per gallon within a few hours of each other. Often, prices dropped early in the morning and increased toward the end of the day, implying that the A.I. software may have been identifying common market-demand signals through the local noise.
The station owners say their systems frequently lower prices to gain volume when there are customers to be won.
“It can be frustrating,” said Erwin Ralan, an electronics-store manager who was filling up at the Tamoil station that week. “Prices usually go up at the end of the day. But when you’re empty and you’re in a rush, there’s not much you can do.”
EndClick here to order this assignment @Essaybay.net. 100% Original.Written from scratch by professional writers.