The war on personalization

Last week, the FTC announced that it has issued information-seeking orders to eight companies offering services related to what it called “surveillance pricing.” From the FTC’s press release:

The orders are aimed at helping the FTC better understand the opaque market for products by third-party intermediaries that claim to use advanced algorithms, artificial intelligence and other technologies, along with personal information about consumers—such as their location, demographics, credit history, and browsing or shopping history—to categorize individuals and set a targeted price for a product or service. The study is aimed at helping the FTC better understand how surveillance pricing is affecting consumers, especially when the pricing is based on surveillance of an individual’s personal characteristics and behavior … “Firms that harvest Americans’ personal data can put people’s privacy at risk. Now firms could be exploiting this vast trove of personal information to charge people higher prices,” said FTC Chair Lina M. Khan.

The FTC goes on to detail the four major areas for which it wants to collect information about these pricing tools:

The eight companies to which the information-gathering orders were issued are Mastercard, Revionics, Bloomreach, JPMorgan Chase, Task Software, PROS, Accenture, and McKinsey & Co.

I find two things noteworthy about the FTC’s announcement. The first is that the commission decided to use the loaded and provocative term “surveillance” to characterize the use of data to “set a targeted price for a product or service.” When it is invoked to describe data collection practices for any use case, this term seems designed to instill a sense of distress, conjuring the notion of a pervasive and sinister Soviet-style domestic spying apparatus. FTC commissioner Melissa Holyoak, in her concurring statement regarding the FTC’s announcement, made a similar point:

First, public statements that accompany the issuance of these orders describe their focus not on targeted or personalized pricing, but on “surveillance pricing.” This term’s negative connotations may suggest that personalized pricing is necessarily a nefarious practice. In my view, we should be careful to use neutral terminology that does not suggest any prejudgment of difficult issues.

The use of a scare term like “surveillance” when describing a practice like dynamic pricing sets a hostile and slanted tone for any discussion of its mechanics. It’s a corrosive approach to framing a discussion. The same was true when the FTC issued its Advance Notice for Potential Rulemaking (ANPR) around the vaguely menacing-sounding practice of “commercial surveillance” back in 2022.

While the semantics are important, the second and more meaningful aspect of the FTC’s announcement related to personalized pricing is that consumer harm isn’t included in the major areas of investigation. Rather, the last bullet point aims to understand how prices change for “surveilled” consumers. Given that personalized pricing necessarily involves exposing different consumers to different price points, it seems that consumer harm is implied as a feature of the practice of personalized pricing — that if prices are being personalized, consumers are worse off for it.

Personalized pricing, which is a form of price discrimination, aims to minimize consumer surplus, or the delta between what a consumer pays for an item and what they might have paid for it, given its value to them. Price discrimination can take three forms, as I describe in Ads in streaming, differential pricing, and the pursuit of ARPU; examples of price discrimination can be found in many commonplace and uncontroversial commercial practices, such as Business Class seating on airplanes and Student or Senior discounts for movie theater tickets.

Part of the FTC’s focus in its investigation of third-party price personalization services seems to pertain to privacy concerns arising from sharing data with third-party services, as well as the potential use of sensitive data for making pricing decisions, both of which are reasonable points of interrogation. But the general thrust of the commission’s concern appears oriented around setting individualized prices for consumers. From the press release (emphasis mine):

The FTC is using its 6(b) authority, which authorizes the Commission to conduct wide-ranging studies that do not have a specific law enforcement purpose, to obtain information from eight firms that advertise their use of AI and other technologies along with historical and real-time customer information to target prices for individual consumers. 

And from the FTC’s blog post about the announcement:

Many consumers today are not actively aware that their devices constantly gather data about them, and that that data can be used to charge them more money for products and services. An age-old practice of targeted pricing is now giving way to a new frontier of surveillance pricing … Advancements in machine learning make it cheaper for these systems to collect and process large volumes of personal data, which can open the door for price changes based on information like your precise location, your shopping habits, or your web browsing history.

In this way, the FTC takes aim specifically at first-degree price discrimination, or perfect price discrimination, which adapts prices to the demand patterns and thresholds of individual consumers. But there is no broad consensus that first-degree price discrimination is categorically hostile or harmful to consumers. In November 2018, the OECD’s Consumer Protection and Competition Committees investigated the effects of first-degree price discrimination using “data analytics and pricing algorithms” on consumer welfare. Its report determines (emphasis mine):

Although the welfare effects of second- and third-degree price discrimination are theoretically indeterminate, first-degree price discrimination unambiguously increases total welfare. By offering discounts to consumers whose marginal value of the good is greater than the good’s marginal cost of production, a firm engaging in first-degree price discrimination expands output and eliminates the deadweight loss associated with market power … Personalized pricing could in some cases enhance competition, increasing both total and consumer welfare. In particular, personalized pricing may intensify competition by allowing firms to target prices to poach their rivals’ customers … In this manner, the competitive use of personalized pricing could lower prices for all consumers, increasing both total welfare and consumer welfare.

The OECD committees also establish that personalized pricing doesn’t violate US antitrust law (emphasis mine):

Finally, U.S. antitrust laws do not prohibit a firm with market or monopoly power from charging any price that the market will bear. Such market prices are integral to well-functioning markets because they provide informative signals about market conditions. This prescription also applies to personalized pricing, which has the potential to ameliorate static welfare losses from monopoly and oligopoly pricing.

Given the generous depiction of first-degree price personalization by the OECD in 2018, why would the FTC characterize the practice as a manifestation of “surveillance” in 2024? One important change in the operating environment across those two periods is the widespread availability of AI tools for achieving price personalization — and, indeed, personalization of many forms, especially with advertising. And my sense is that the proliferation of AI-based personalization tools — used to optimize not only prices but also in-product content and user experience elements — will attract continued scrutiny given the automated and largely opaque nature of the process.

Personalized advertising may have been the first such practice to draw ire but others, such as pricing and video game optimization, have also attracted criticism. I believe that AI-enabled personalization, across use cases such as pricing, advertising, and content curation, is the next technology-related moral panic. Personalization practices that have existed for, in some cases, decades can be made to seem malevolent when enacted through AI-enabled tools. And that depiction is all the more potent when couched in loaded, rousing, and exhortative terminology like “surveillance.”