Algorithms and pricing: The underlying logic of "big data killing" 丨 Wei Yan Yi

When we have become accustomed to the convenient consumption experience brought to me by digital technology, concerns about data security and control of big data algorithms have gradually emerged. What is the current approach of global businesses to these issues? What options do we have at the moment? This article aims to sort out the changes in pricing models and try to disassemble the underlying logic of big data in our definition today.

- Editor

One

In September 2000, Amazon.com conducted a dynamic pricing experiment in which DVDs were sold to customers at different prices (up to a 40% difference) based on their purchase history. When news of the experiment came to light, consumer privacy groups offered harsh criticism. Amazon also publicly apologized and returned the money to 6896 customers.

The Washington Post has since tracked the matter in a report titled "On the Web Price TagsBlur: What You Pay Could Depend on Who You Are": "While the dynamic pricing of DVDs has received more attention in the media, Amazon's dynamic pricing was actually first discovered in May 2000. At the time it was found offering discounts of over 20% to some customers on a popular MP3 player. In addition to dynamic pricing, businesses use consumer profile data to target ads and make product recommendations. In fact, customized ads that use consumer profile data sell for ten times as much as untargeted ads. ”

More than 20 years later, how is this dynamic pricing progressing? With the shift of the Internet from PC to mobile, the problem of dynamic pricing has not been solved, on the contrary, personalized dynamic pricing has emerged.

Worldwide, while privacy concerns are getting more attention and there are more laws restricting merchants from using personal information, consumer concerns have not been completely eliminated. Coupled with the report of "big data killing" coming from time to time in the media, many consumers will wonder if I have been treated specially and "discriminated" by the algorithm? What should the law do?

Two

Typically, an algorithm is "a process or set of rules to follow in a calculation or other problem-solving operation." In the context of revenue and revenue management, algorithms maximize revenue and profits by calculating the optimal distribution of goods and adjusting prices accordingly, a strategy that typically uses algorithms or software to determine prices is known as "algorithmic pricing." Historically, algorithmic pricing didn't start with the internet, it didn't originate with airlines.

Not only is the initial investment of airlines huge, but each flight has a huge expenditure, so a good pricing system is a prerequisite for the airline to operate well. Airlines have long aimed to determine seat prices in a way that maximizes revenue and has resulted in a unique Revenue & PricingManagement System. Taking into account the huge differences in the willingness of different customer groups to pay, airlines differentiate their fares through restrictive fare rules, making each customer base tend to a different fare class. These rules include: the price is different at different times in the same location, and the corresponding refund and change rules are confirmed according to the fare. The revenue management system restricts the supply of low fares for high-demand flights in order to reserve seats for future higher revenues; it also sets seat inventory levels to maximize total flight revenue while balancing unsold seats at departure (waste of inventory) with not having enough seats to meet high-revenue last-minute demand (demand overflow).

American Airlines' first automated revenue management system introduced in the 1980s was hailed as "the most important technological development in transportation management" by its then-chairman and CEO, Robert Clandall, who estimated that sustained $500 million a year in revenue benefited from its investment in revenue management and overbooking systems. The Initial Revenue Management System uses a heuristic approach to optimize seat inventory for each segment individually. Algorithmic pricing was seen as a key driver of business success in terms of increasing profits, and after being successfully used by American Airlines, the hotel industry, similar to the airline industry, quickly adopted a revenue management system.

Three

Today, algorithm pricing has expanded to many industries, but the most fully used is the Internet company, the reason is that this industry can access more user information, it can track user behavior information at near zero cost to provide users with the right product and determine the right price. As the economist Krugman pointed out in a New York Times op-ed in 2000: "Dynamic pricing is a new version of an old practice, price discrimination. It uses the potential consumer's electronic fingerprints — his previous purchase history, his address, perhaps other websites he visits — to gauge how likely he is to back off in the face of high prices. If the user seems price-sensitive, he gets a bargain; if he doesn't, he pays a high price. ”

The same is price discrimination, why is the airline's revenue management system highly praised, while the Internet company's algorithmic pricing is mixed? The reason is that Krugman said that Internet companies can obtain relevant information about individual users, whereas previously the airline's revenue management system did not have the identity information of specific guests. It is also true that many people have called for that major companies should be prohibited from collecting user information? So as to ensure that consumers are not discriminated against by algorithms?

It should be noted that not all consumers object to the collection of information by websites or apps. As Baidu CEO Robin Li said at the 2018 China Development Forum that "Chinese users are often willing to exchange privacy for convenient services", of course, Robin Li's original words are like this:

At the same time, some of the user's personal data can actually help the connected enterprise to provide better services or products for it. For example, the user's habits on e-commerce and shopping websites, the categories of concern, and so on, help the website to provide users with more intimate and efficient services. Under the premise of privacy protection, Chinese consumers are often willing to authorize the use of certain personal data in exchange for more convenient services. Therefore, we need to find a better balance between ensuring the security of user information and using user data to provide better services. Of course, all this must follow a certain principle, on the basis of protecting the rights and interests of user data, use this data to benefit everyone. ”

Although many people don't like to hear this sentence of Robin Li, in fact, it is true. Although consumers value privacy, privacy is not everything. Back in 2005, Professor Alessandro Acquisti of Carnegie Mellon University and Professor Hal R. Varian of the University of California, Berkeley, noted in a paper titled "Adjusting Prices Based on Purchase History" that they would disclose personal information as long as there was sufficient incentive. It is also based on this that they also suggest in this paper that there is a competitive advantage in industries where there is a large initial investment in information technology, such as travel and online shopping, by enhancing personalized services, because the marginal cost of these services is very low for sellers. As long as the price is right, this practice can lead to increased customer loyalty and increased overall welfare.

Many times consumers are willing to disclose information online, not only because there is sufficient incentive, but also because retaining information will bring benefits to individuals. In 2006, in his article Online Disclosure: Motivations and Metrics, Xu Qilong of the National University of Singapore summarized the benefits of disclosing personal information, which bring four external benefits: saving money, saving time, self-improvement and social adaptation, as well as three intrinsic benefits: pleasure, novelty, and altruism. Indeed, the reason why users like to shop online is because the platform can recommend relevant products and services to you according to the user's behavior, if you do not leave a trace every time you go online, is it too boring?

Four

However, even though the personalized service of the Internet has brought many conveniences to users, some people still have lingering feelings about it, especially whenever they see the "big data killing" reported in the media. Will I be the "cooked" who was "killed" by big data?

But in my opinion, this worry is superfluous. True personalized pricing is difficult to achieve for a variety of reasons. The first is that it stems from technology, and a good algorithm is not easy. A phrase that programmers often mention is, "Garbage in, garbage out", and if you enter wrong, meaningless data into a computer system, the computer will naturally output incorrect and meaningless results. Matching user habits to price is easy to say, but not easy to operate. For example, major airlines and hotels often adjust the rights and interests of members, on the bright side, according to changes in the situation, at worst, the preset standards cannot catch up with the changes, resulting in the unattainable goals. A typical example is Marriott, whose CEO said in an interview with the media about the change of membership plan, in public that "members have taken too much", which means that the company has lost. A static membership system can produce all kinds of flaws, not to mention dynamic personalized pricing.

The second is management costs. The potential revenue and profit gains, along with the assumption of low adaptation costs, make algorithmic pricing very attractive to many companies. But many times, this dynamic pricing will have very large management costs within the company, such as some scholars use the "myth of no-cost price changes" to describe the dynamic pricing mechanism, because it ignores the need for an in-depth assessment of customers, supply chains and company structure before any new pricing policies are introduced. At the same time, within the organization, this pricing mechanism also incurs the physical costs associated with the installation and maintenance of the IT infrastructure. In this regard, Ahmad Faruqui and Sanem Sergici, an example of pricing mechanisms in the U.S. electricity market, illustrate that management costs also limit the use of pricing mechanisms. Since the 2000-2001 energy crisis in the western United States, improving demand response in the electricity market has attracted much attention. Economists suggest that one of the best ways to provide a demand response from the electricity market is to pass on wholesale energy costs to retail customers, which can be achieved by having retail prices change completely or partially dynamically. But the plan would require changes to metering infrastructure, which could cost up to $40 billion for the entire United States. Because of the high cost, this one ended up almost stillborn.

The third is reputation costs. Although many studies have shown that personalized pricing, that is, primary price discrimination, contributes to the overall welfare of society, this differentiated pricing approach is still difficult for consumers to accept. It is also in this way that Amazon was pointed out to implement the experiment of personalized pricing, and immediately refunded the amount of overcharged money, which is to establish an image in the minds of consumers.

Much of the literature points out that algorithmic pricing literature suggests that price framework strategies reduce the perception of fairness, thereby reducing customer trust in the company. Thinking about it, why should the same goods and services have to be different from others, just because I have a high income? It is also true that many companies do not implement price discrimination in the form of personalized pricing, but by issuing coupons. Particularly targeted coupons displayed at discounts are an effective framework strategy that can mask the sense of unfairness that comes with personalized pricing.

It is precisely because the cost of algorithmic pricing is so high that some companies simply give up this idea, but directly implement price discrimination in the form of membership fees, and then do not play tricks on the price of goods and services. For example, the recent market openers who have reached out to the market, according to the latest financial report, its membership fee income is 3.877 billion US dollars, and its annual profit is 5.079 billion US dollars, and the membership fee income accounts for about 76.33% of the total profit, and in 2009, this proportion was as high as 86.27%.

Five

In the face of changes in pricing methods, the law has also made corresponding preparations. Article 18 of the E-commerce Law, which came into effect in 2019, stipulates that "where e-commerce operators provide consumers with search results for goods or services based on their interests, hobbies, consumption habits, and other characteristics, they shall at the same time provide the consumer with options that are not specific to their personal characteristics, respecting and equally protecting the legitimate rights and interests of consumers." Article 24 of the Personal Information Protection Law also stipulates: "Personal information processors using personal information to make automated decisions shall ensure the transparency of the decision-making and the results are fair and just, and must not implement unreasonable differential treatment of individuals in terms of transaction prices and other transaction conditions." The Cyberspace Administration of China and the State Administration for Market Regulation have also listed algorithm discrimination as the target of regulation in a series of departmental rules and normative documents.

This change in the law is certainly necessary, but I don't think consumers have to worry too much about it. In a competitive market, competitive pressures and reputational mechanisms can limit companies from abusing personalized pricing, and even produce mechanisms that are completely different from personalized pricing, just like market openers, "Don't make money, make a friend." ”

(The author is a researcher at Shanghai Academy of Finance and Law)