E-commerce has clear economic benefits. It has expanded consumer choices, reduced the cost of search and comparison, and allowed companies to reach markets they could not previously access. With artificial intelligence, platforms are better able to analyze demand, predict tastes, suggest products, and direct advertising to people most likely to buy, rather than spending money on general advertising campaigns that the seller does not know who will respond to.
But behind this efficiency lies a completely different economic equation than that faced by the traditional store. The store owner chooses the goods before the season, buys them, finances them, ships them, and clears them through customs, then bears taxes, fees, rent, employee salaries, electricity, and equipment. And after all that, they wait for the consumer to enter through the door.
If the merchant misjudges the color, size, fashion, or demand volume, the goods remain on the shelf. The capital that was supposed to circulate and turn into sales becomes stagnant inventory. Then the discounts begin, and the merchant may have to sell at the end of the season at a small margin, or even at a loss, to recover some of the liquidity.
The global e-commerce platform can operate with a different equation. It can display thousands of products, test demand on them digitally, and know which products attract interest before they reach the consumer, then direct the desired goods directly to them. Thanks to the enormous scale of its operations, it can distribute the cost of technology, storage, marketing, and shipping over millions of transactions and dozens of markets.
In Jordan, the issue of taxes and customs is added to this. Commercially imported clothing passes through a customs and tax system where tariffs vary according to the type, classification, and origin of the goods, and the applicable trade agreements. The local merchant also bears the rest of the costs of operating within the Jordanian economy.
But the real difference between the two models may not be in taxes, rent, or even inventory, but in information. The traditional store knows a lot about the goods and relatively little about the person who will come to buy them. The digital platform may know what the consumer searched for yesterday, what they bought last month, the products they viewed but did not buy, the time they spent viewing a particular product, the prices they usually pay, their location, and dozens or perhaps hundreds of other signals resulting from their digital behavior.
This information was primarily used for advertising: if you are interested in sports, you are shown a sports shoe, and if you search for a trip, hotel and flight offers start appearing in front of you. But artificial intelligence allows for a more complex stage. Instead of using data only to find out what you want, it can be used to try to find out how much you are willing to pay.
Here we approach an old economic concept, first-degree price discrimination, which I will explain with an example: suppose three people want to buy the same item. The first is willing to pay 50 dinars, the second 40, and the third 30. If the advertised price is 30 dinars for everyone, the first and second get a benefit because they were willing to pay more than the actual price. This benefit is called consumer surplus by economists. If you are willing to pay 50 dinars for a shirt and buy it for 30, you have a consumer surplus of 20 dinars.
But what happens if the algorithm can know, or at least predict with a high degree of accuracy, that you are willing to pay 50, from the seller's perspective, the ideal price for you is closer to 50, while it may be 40 for another consumer and 30 for a third. The closer the seller gets to knowing each person's maximum willingness to pay, the closer they theoretically get to first-degree price discrimination, and the greater the portion of consumer surplus that goes to the seller. That is, what was a theoretical assumption difficult to apply in economics textbooks has become capable of being approximated by artificial intelligence and big data.
This personalized pricing is not necessarily harmful in all cases. A company may use the information to offer a lower price to a price-sensitive consumer, who will buy an item they would not have bought at a uniform price. Price discrimination, under certain conditions, can expand the market and increase sales. But the problem starts when the consumer does not know that the price they see may not be the same as what others see, and when personal and behavioral information is used to estimate the maximum amount that can be taken from them, without knowing the size or how it was collected.
What if the same idea moves to banks and loans? Here the issue becomes more important. Banks already use a risk-based pricing system. A customer with a higher likelihood of default may pay a higher interest rate than a customer with a strong credit history and stable income. The difference in price here reflects the difference in the cost of risk. But artificial intelligence can, in principle, allow the bank to answer a completely different question, which is: what is the highest interest rate this customer would be willing to accept?
Suppose the customer's risk, the cost of funds, and the normal profit margin justify a loan at 7% interest. But the bank can know through data that the customer's salary goes into the same bank, that they have urgent financial obligations, that their balance is declining rapidly, that their transaction pattern indicates a need for liquidity, that they do not compare bank offers much, and that they have previously accepted credit offers quickly. If the algorithm can deduce that they would accept a loan at 9.5%, should the bank offer them 7% or 9.5%?
If the increase reflects a higher likelihood of default, it is risk pricing. But if the customer is no different in risk from others, but pays more just because the algorithm discovered that they are more willing or compelled to pay, we are facing something completely different, which is pricing for willingness to pay. Here, banking pricing itself approaches the idea of first-degree price discrimination.
Banks and financial institutions have more sensitive data than that available to an e-commerce store. They may know salary flows, balances, monthly spending, card usage, bill payment regularity, existing loans, transfers, and cash withdrawals. As the use of alternative data expands, other signals can enter the evaluation models.
The better the bank knows the customer, the better it can estimate their risk; but it may also become better at estimating their negotiating weakness. If this knowledge is used not to estimate the likelihood of repayment but to estimate the maximum interest rate the customer would accept, the data turns from a tool for measuring risk into a tool for extracting consumer surplus, or what can be called, in colloquial language, consumer exploitation.
This is where the role of economic policy, competition laws, data, and consumer protection comes in. The goal is not to resist artificial intelligence, but to define the boundaries between allocation that increases efficiency and allocation that extracts consumer surplus based on information that the consumer does not have the same power to counter. Here, too, begins the next major economic challenge.
