Some subscribers recently received a heads-up that they're on the hook for a new rate "set by an algorithm using your personal data." We asked a UVA expert what that might mean.
Photograph by Evy Mages.
If recent events have not compelled you to cancel your Washington Post subscription, then you might have been in for sticker shock at the dawn of your latest billing cycle. Many readers have been notified via email that their subscription rates are set to increase. Nestled at the bottom of these emails, you’ll find an asterisk and the following: “This price was set by an algorithm using your personal data.”
The Post‘s use of algorithmic pricing is not surprising, given the newspaper’s recent fixation on artificial intelligence—consider its AI-powered search engine and robot-led podcast. When we asked the Post for comment on its algorithmic pricing mechanisms, a spokesperson directed us to a blog post from the publication’s engineering team. The article explains how an AI-driven “smart metering model” determines the number of free articles both anonymous users (who are not registered on the Post‘s website) and registered users (who have free online accounts but no paid subscription) can access before a paywall pops up. But it doesn’t touch specifically on how the Post uses subscriber information to determine pricing.
For some insight on how companies might mine reader data in order to maximize profit, we reached out to Luca Cian, a professor at the University of Virginia’s Darden School of Business. Though Cian does not have firsthand knowledge of how the Post‘s algorithmic pricing model works, he says that many such models rely on user demographics and location to determine how much they might be willing to pay for a product.
Companies have long been using geographic information to dictate pricing. In 2015, ProPublica reporters discovered that the Princeton Review was charging more for the same SAT tutoring package in areas with higher Asian populations. “They call it the ‘tiger mom tax’ because they thought, ‘OK, that population most probably will really be focused on things like SAT product packages,'” Cian says. But the advent of real-time AI models has enabled companies to engage with individual user data more intricately. Instacart recently killed an algorithmic pricing model that allowed grocery stores to charge certain shoppers as much as $2.56 more for the same item. Post owner Jeff Bezos’ Amazon came under fire last year when the retailer’s dynamic pricing mechanism reportedly charged local school districts vastly different prices for the same supplies, sometimes even on the same day.
A Federal Trade Commission study published early last year found that users’ browser histories and location data are often used to influence the prices they see while online shopping, but the agency hasn’t made moves to regulate the practice. Some action has been taken at the state level: In November, New York adopted a law requiring companies to disclose any use of algorithmic pricing to consumers. So far, California has the nation’s most comprehensive laws on the matter; a series of related antitrust amendments took effect at the start of this year, forbidding competing companies from using shared algorithms to set prices and coercing customers to accept the prices established by such algorithms. Locally, Maryland Governor Wes Moore has introduced legislation to prevent grocery stores from using consumer data to charge individualized prices.
When the Post‘s algorithm evaluates the ideal subscription price to charge for a reader, Cian says, the company “can calculate in real time a high level of complexity based on massive data they acquire throughout the year, based on all the data that they know about their subscribers and when they did or did not renew their subscription.” Your rate might come down to assumptions that the algorithm makes about your financial status based on how you access digital articles. “If you use an Apple product, usually people increase prices because they assume that if you have an iPhone, you may have a higher income than if you have an Android,” he adds. “They know exactly from your IP address where you are reading most of the time, so they can access through Zillow how much is the average cost of a house in that area [and] probably infer really quickly your income.”
Readers’ usage of the Post‘s services might also play a role in how much they’re charged. “Users that read a lot may need to be paying more because they actually use more of the services—you can say, probably, they value our services more so we can charge them a little bit more,” Cian says. Whereas, for subscribers who don’t read as many articles to begin with, “maybe you don’t want to affect their pricing too much, because otherwise you stand to lose them.”
What really interests Cian, who has published research exploring how audiences tend to have less trust in media outlets that are transparent about their AI use, is the fact that the Post disclosed its use of algorithmic pricing at all. “If you ask people [whether they] want transparency on what’s behind your pricing strategy, people say ‘yes,'” he says. “But what we found in my research is a paradox, in the sense that people think that they want to know, but once they know, the reaction is worse than not knowing.”
Is it possible for Post subscribers to limit how much personal information the company can access? Yes and no. “Most of the time, we give up data when we accept the user agreement,” Cian says. Using a bare-bones “dumb phone” as opposed to a smartphone to read articles online can offer some anonymity, as can visiting the website using a VPN browser that hides your location. “It’s not impossible, but it requires a lot of effort and I would say probably 0.5 percent of the population does it,” Cian adds. “We are in an age and time where we might need to assume that there is very little privacy left.”
