The Price Is Watching You: How Algorithms Learned to Charge More
For most of modern commerce, a price was at least visible. A gallon of milk, an airline ticket or an apartment had a listed cost. Customers might dislike the number, but they could compare it with competing offers and decide whether to buy. Sellers lacked the ability to recalculate the price continuously based on a shopper’s browsing history, location, device, urgency or estimated willingness to pay.
Algorithms are changing that relationship. Companies can now adjust prices in real time, monitor competitors automatically and combine purchasing behavior with extensive personal data. The result can be efficient when it helps businesses respond to inventory, demand or changing costs. It becomes more troubling when the same technology is used to estimate the highest price a particular customer will tolerate or to reduce the incentive for competitors to undercut one another.
The Federal Trade Commission calls one version of this practice “surveillance pricing”: using data about consumers’ characteristics and behavior to set or target prices. Its investigation found that intermediaries can incorporate information such as browsing activity, precise location, purchase history and other personal characteristics into pricing tools. The price is no longer based only on what the product costs or what the market will bear. It may increasingly be based on what an algorithm believes you will bear.
The Algorithm Knows More Than the Customer
Dynamic pricing is not inherently new or unlawful. Airlines, hotels and utilities have varied prices for decades according to demand, availability and timing. A hotel room may cost more during a major event, while electricity may cost more during periods of peak use. Modern pricing technology can go much further.
A seller may know which products a customer viewed, how long the customer remained on a page, whether the person has searched several times, what kind of device is being used and whether the buyer appears to be in a hurry. An intermediary can combine that information with data purchased from other sources and recommend a price, promotion or financing offer. The FTC has investigated whether these tools allow companies to charge different consumers different prices for the same goods and services. The agency has warned that surveillance pricing may affect purchases ranging from groceries and vehicles to housing.
Traditional price discrimination was often easy to see. A student discount, senior rate or weekend promotion came with understandable conditions. Algorithmic personalization can be invisible. Two shoppers may see different offers without knowing the difference exists or which personal information produced it. That opacity changes bargaining power. The seller may have an extensive profile of the consumer, while the consumer sees only one number and has no way to determine whether someone else received a better one.
Amazon Allegedly Tested How High Prices Could Go
Amazon’s pricing practices demonstrate how an algorithm can influence prices beyond a single website. In an antitrust complaint filed by the FTC and a coalition of states, regulators alleged that Amazon used a secret algorithm known as Project Nessie to identify products for which it could raise prices while predicting whether competing retailers would follow. According to the complaint, when competitors matched the higher Amazon price, Amazon could keep its price elevated. If rivals did not follow, the system could reverse the increase.
The regulators alleged that Project Nessie generated more than $1 billion in additional profit for Amazon. These are allegations in litigation rather than a final judicial finding that every challenged act occurred as characterized. Amazon has disputed the government’s broader antitrust case and has said the program was discontinued. The significance of the allegation reaches beyond Amazon customers. A platform with enormous market visibility may be able to test a higher price and watch whether competitors respond. When other retailers use Amazon as a pricing reference, an increase on the platform can influence prices elsewhere.
The conventional image of competition assumes that when one retailer raises a price, another will undercut it to attract the customer. The FTC’s theory is that Amazon’s market power and automated pricing systems could produce the opposite result: competitors follow the increase, allowing the higher price to spread. No explicit phone call among retail executives would be necessary. Software can observe, respond and reinforce the change in seconds.
Price Fixing Does Not Become Legal When Software Does It
Federal antitrust law prohibits competitors from agreeing to fix prices. The legal analysis becomes more difficult when companies do not communicate directly but feed sensitive information into the same pricing platform. An algorithm does not need to announce, “Let us all charge more.” It can recommend prices based on confidential information supplied by several competitors. When participants repeatedly accept those recommendations, the software may reduce the independent decision-making that competition requires.
The Justice Department has said antitrust law does not become obsolete merely because companies use technology to coordinate. Its rental-housing case against RealPage alleged that landlords shared nonpublic data about rents, vacancies and future availability, which RealPage’s software used to generate pricing recommendations.
The central issue is not whether an equation or artificial-intelligence system can itself form a criminal intention. It is whether businesses use the technology to share competitively sensitive information, align prices or avoid the independent rivalry that should benefit customers. An algorithm can be a legitimate analytical tool. It can also become the conference room in which coordination occurs.
Renters Became Test Subjects for Algorithmic Pricing
Housing demonstrates how algorithmic pricing can affect a necessity consumers cannot easily refuse. A renter cannot postpone housing indefinitely because the price appears unfavorable. Moving is expensive, vacancies may be limited and landlords frequently require application fees, deposits, credit checks and income above a specified multiple of the rent.
The Justice Department sued RealPage in August 2024, alleging that the company’s revenue-management software allowed competing landlords to pool sensitive information and align rental pricing. The government later expanded the case to include several large landlords.
RealPage has maintained that its software helps property owners analyze markets and operate more efficiently. In November 2025, the Justice Department filed a proposed settlement that would restrict RealPage’s use of competitively sensitive information and certain pricing practices. RealPage did not admit liability in agreeing to the proposed resolution.
The case illustrates the unusual power of shared data. A landlord traditionally had to estimate what nearby competitors were charging from public listings and local experience. A centralized platform can collect daily, nonpublic information from a large group of properties, including actual lease terms and future availability. That can produce more accurate forecasts. It can also tell landlords that they may be able to hold units vacant rather than lower rents, particularly when many nearby competitors receive similar advice. A market cannot function competitively when every major participant is effectively looking at the same private playbook.
The Consumer May Be Negotiating Against an Entire Market
Algorithms can make companies faster and more consistent. They can also eliminate the small inefficiencies that once benefited consumers. A local manager might once reduce rent because several apartments were empty or discount a product to clear inventory. Centralized software can discourage that deviation by telling the manager that a higher price will maximize revenue across a portfolio. From the company’s perspective, that is disciplined management. From the consumer’s perspective, it can mean that there is no longer a genuinely independent seller willing to make a better deal.
The concern becomes greater when the same software provider serves a significant portion of an industry. Every company may remain legally separate, yet their decisions become more uniform because they rely on similar data, assumptions and recommendations. Competition depends on independent judgment. When algorithms standardize that judgment, markets can begin to behave as though coordination exists even when no executive has signed a conventional price-fixing agreement.
Healthcare Pricing Is Already Too Opaque
Healthcare is particularly vulnerable to algorithmic power because patients rarely behave like ordinary shoppers. A person experiencing chest pain does not compare emergency-room prices. A patient may not know whether a clinician, anesthesiologist or laboratory is in the insurance network. Even when prices are disclosed, they can depend on negotiated rates, deductibles, coding decisions and whether an insurer approves the service.
Algorithms are used throughout healthcare to process claims, identify fraud, assess medical necessity and manage payment. These tools can improve efficiency, but they also affect whether care is approved, how bills are categorized and which claims receive additional review. The strongest documented concerns in healthcare frequently involve automated coverage and claims decisions rather than a single industrywide algorithm that directly sets all medical prices. It would therefore be inaccurate to blame healthcare inflation primarily on artificial intelligence.
The broader power imbalance is still similar. Insurers and providers possess sophisticated data systems, actuarial models and negotiated-price information. Patients often receive a bill they cannot explain after the service has already occurred. When an algorithm denies, delays or reprices a claim, the patient may struggle to discover what rule was applied or how to challenge it. Automation can make a flawed decision faster without making it fairer.
Personalized Pricing Can Punish Urgency
One of the most troubling possibilities is that pricing systems may identify when a customer has limited alternatives. A person repeatedly searching for a last-minute flight may be traveling for a funeral. Someone researching an emergency plumbing repair cannot wait several weeks for a better offer. A patient seeking medication may have no practical substitute. A company that detects urgency can use it to serve the customer more quickly. It can also interpret urgency as a greater willingness to pay.
The ethical problem is not merely that prices change. It is that the customer may be charged more because the system has inferred vulnerability. Markets have always charged premiums for speed and scarcity. Algorithmic surveillance makes it possible to personalize that premium according to the individual’s behavior, location and perceived desperation. A price determined partly by a person’s inability to walk away is not the same as a price generated through open competition.
Discounts Can Be as Manipulative as Price Increases
Surveillance pricing does not always appear as a higher listed price. It can also operate through selective discounts. A company may keep the public price high while offering a coupon only to customers its model believes need an incentive. Loyal customers who are likely to buy anyway may receive no discount. New or price-sensitive consumers may receive a more attractive offer. From a marketing perspective, this is efficient. The company avoids giving away margin to customers willing to pay the full price. For consumers, the practice means loyalty may be penalized. A frequent buyer can pay more precisely because the algorithm knows that person is unlikely to leave.
Personalized promotions can also create discriminatory effects even when protected characteristics are not explicitly entered into the system. Location, shopping patterns, income estimates and other proxy variables may correlate with race, age, disability or economic status. The algorithm does not need to know someone’s identity in ordinary human terms. It only needs enough information to predict behavior.
Employers Have Their Own Information Advantage
The same imbalance appears in the labor market. Employers increasingly use automated systems to advertise jobs, screen résumés, rank applicants, administer assessments and evaluate employees. The Equal Employment Opportunity Commission has warned that these systems can violate federal discrimination laws when their design or use disadvantages protected groups.
The worker usually knows little about the model. An applicant may never learn which résumé phrase caused rejection, whether an employment gap was penalized or whether a disability-related response affected an automated assessment. Companies can also acquire extensive compensation data from payroll providers, consultants and employment platforms. A job seeker usually has much less information about what a company has paid comparable workers, how urgently the position must be filled or how much budget remains available.
The result is an asymmetric negotiation. The employer can analyze thousands of salaries and applicants, while the individual may negotiate based on a few public job listings and conversations with colleagues. Algorithms do not create that imbalance, but they can deepen it by allowing employers to process market information at a scale no individual worker can match.
A Hiring Model Can Reproduce Yesterday’s Bias
An automated system is often described as objective because it applies the same mathematical process to every applicant. That does not mean the process is neutral. If a model is trained on a company’s past hiring decisions, it may learn patterns created by earlier discrimination. If employees historically selected for leadership came disproportionately from one demographic group, the system may identify characteristics associated with that group and treat them as signals of future success.
The EEOC has noted that software can learn discriminatory preferences by observing past decisions even when an employer does not expressly instruct the system to discriminate. Automated assessments may also disadvantage people with disabilities when a test measures an impairment rather than the skill required for the job. A video system analyzing facial movements, voice or eye contact can misinterpret a qualified applicant whose disability affects those characteristics. The EEOC and Justice Department have specifically warned employers about disability discrimination involving algorithmic hiring tools. Efficiency does not excuse the result. An employer remains responsible for discriminatory employment decisions even when a third-party vendor supplied the software.
Algorithms Do Not Cause All Inflation
It is tempting to attribute rising prices broadly to algorithmic manipulation. That conclusion goes beyond the evidence. Inflation can result from supply disruptions, energy prices, labor costs, housing shortages, government policy, consumer demand and many other forces. An algorithm cannot make a scarce apartment abundant or lower the cost of producing a product.
Pricing technology can influence how quickly businesses respond to inflation and how completely they pass higher costs to consumers. It may also help companies recognize that customers will tolerate a price increase even when the company’s own costs have not risen proportionately.
The economic concern is therefore not that algorithms created every price increase. It is that they can make price increases easier to coordinate, more precisely targeted and harder for consumers to escape. Technology can turn a temporary shortage into an opportunity to test consumers’ maximum tolerance and retain part of the increase after the original pressure subsides.
Market Power Determines How Dangerous the Tool Becomes
A pricing algorithm used by one small retailer in a highly competitive market has limited power. If the seller raises prices too aggressively, customers can leave. The danger grows when the company controls a major platform, possesses unusually rich data or provides pricing software to many supposed competitors. Customers may have few practical alternatives, while rival businesses may depend on the same marketplace or analytical system. The FTC’s Amazon case alleges that the company used interlocking practices to maintain monopoly power in online retail markets, including measures that discouraged sellers and competitors from offering lower prices elsewhere. Amazon disputes the allegations.
This is why the same software can have different consequences depending on who controls it. A recommendation engine is not inherently anticompetitive. A recommendation engine operated by a dominant gatekeeper can shape the behavior of an entire market. Algorithms magnify existing power. They do not distribute it evenly.
Consumers Cannot Compare Prices They Cannot See
Competition assumes that buyers can observe alternatives. Personalized pricing weakens that assumption because each customer may see only one version of the market. A consumer cannot know whether a price is competitive when other people are receiving different prices based on criteria the company will not disclose.
Browser cookies can be deleted, and shoppers can compare devices or use private browsing modes, but individuals cannot realistically audit the entire data ecosystem behind an offer. Information may come from loyalty programs, data brokers, location histories, prior purchases and inferences that cannot be corrected because the consumer does not know they exist.
This is not ordinary haggling. In a negotiation, both parties understand that a negotiation is occurring. Surveillance pricing can conduct the negotiation silently, before the customer realizes a price has been personalized. The company arrives with a behavioral profile. The customer arrives with a shopping cart.
Transparency Alone Will Not Restore Competition
Requiring businesses to disclose that prices may vary would help consumers understand the practice, but a disclaimer does not solve the underlying imbalance. A notice stating that “prices may be personalized” does not reveal which data were used, how much the price changed or whether opting out is realistically possible. Consumers may still need the product, apartment, flight or medical service.
Meaningful protection could require companies to disclose when individualized data affect a price, identify the major categories of information used and provide a way to view a nonpersonalized offer. Regulators may also need access to the models and data necessary to determine whether pricing systems facilitate coordination or discrimination.
Trade-secret protection complicates this oversight. Companies have legitimate reasons not to reveal proprietary software publicly. That cannot become a shield preventing regulators from testing whether a system breaks the law. The public does not need every line of code. It needs confidence that someone independent is examining what the code does.
The Law Must Focus on Outcomes, Not Vocabulary
Companies frequently describe pricing tools as optimization, revenue management or artificial intelligence. None of those labels determines whether the conduct is legal. An algorithm that forecasts demand may be lawful and useful. Software that allows competitors to share sensitive information and coordinate prices may violate antitrust law even when the coordination is described as a recommendation. The Justice Department’s RealPage action makes this principle explicit: competitors cannot avoid antitrust scrutiny by outsourcing pricing decisions to a common algorithm.
Enforcement will remain difficult because algorithms change rapidly and may produce outcomes through complex interactions rather than a simple instruction to raise prices. Regulators need technical expertise, access to data and the ability to examine how recommendations are implemented in practice. A system should be judged by the market behavior it produces, not by whether a human executive typed the final number.
Convenience Has Become the Cover Story
Companies promote algorithms as tools that improve speed, consistency and relevance. Those benefits are real. A hotel can manage thousands of room rates, a retailer can respond to inventory shortages and an employer can process applications more quickly. Consumers may receive useful recommendations and discounts that would not otherwise exist. Convenience should not prevent scrutiny of who captures the benefit.
An algorithm that reduces a company’s costs could produce lower consumer prices. It could also allow the company to preserve the savings as profit while charging each buyer the highest predicted amount. A hiring tool could reduce bias or reproduce it at a larger scale. A rental platform could help landlords understand demand or discourage them from competing on price. Technology does not determine which outcome occurs. Incentives and market power do.
The New Price Tag Is a Prediction
The old price tag described what a seller wanted for a product. The new price tag may describe what a machine believes about the buyer. It may predict income, urgency, loyalty, alternatives and sensitivity to price. It may know what nearby competitors charge and whether they are likely to follow an increase. It can update that prediction more quickly than a consumer can open another browser tab.
This does not mean every fluctuating price is evidence of manipulation. Dynamic pricing can allocate scarce inventory and help markets respond efficiently. The problem begins when consumers cannot determine whether they are seeing a market price or a private assessment of their vulnerability. Algorithms have given companies extraordinary power to observe the people on the other side of a transaction. Consumers have received little comparable power to observe the algorithm. That imbalance is becoming one of the defining economic questions of the digital marketplace.
The future of pricing will not be decided merely by whether companies use artificial intelligence. It will be decided by whether regulators allow the technology to strengthen competition—or permit it to simulate competition while quietly teaching every seller how to charge more.
