Your boss wants to use surveillance data to cut your wages
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
What industry calls "personalized pricing" is really surveillance pricing: using digital tools' flexibility to change the price for each user, and using surveillance data to guess the worst price you'll accept:
At root, surveillance pricing allows companies to revalue both your savings and your labor. If you get charged $2 for something I only pay $1 for, the seller is essentially reaching into your bank account and revaluing the dollars in it at 50 cents apiece. If you get paid $1 for a job that I make $2 for, then the boss is valuing your labor at 50% of my labor:
Surveillance pricing is a key part of enshittification, relying on three of the key enshittificatory factors that have transformed this era into the enshittocene:
I. Monopoly: Surveillance pricing is undesirable to both workers and buyers, so in a competitive market, surveillance pricing would drive labor and consumption to non-surveilling rivals:
II. Regulatory capture: Surveillance pricing only exists because of weak regulation and weak enforcement of existing regulations. To engage in surveillance pricing, a company must first put you under surveillance, something that is only possible in the absence of effective privacy law.
In the USA, privacy law hasn't been updated since Congress passed a law in 1988 that banned video-store clerks from disclosing your VHS rentals:
In the EU, the strong privacy provisions in the GDPR have been neutralized by US tech giants who fly an Irish flag of convenience. Ireland attracts these companies by allowing them to evade their taxes, but it can only keep these companies by allowing them to break any law that gets in their way, because if Meta can pretend to be Irish this week, it could pretend to be Maltese (or Cypriot, Luxembourgeois, or Dutch) next week:
What's more, competition laws in the EU and the USA ban surveillance pricing, but a half-century of lax competition law enforcement has allowed companies to routinely engage in the "unfair and deceptive methods of competition" banned in both territories.
III. Twiddling: "Twiddling" is my word for the way that digitized businesses can use computers' flexibility to alter their prices, offers, and other fundamentals on a per-user, per-session basis. It's not enough to spy on users: to engage in surveillance pricing, you have to be able to mobilize that surveillance data from instant to instant, changing the prices for every user. This can only be done once a business has been digitized:
https://pluralistic.net/2023/02/19/twiddler/
Combine monopoly, weak privacy law, weak competition law, and digitization, and you don't just make surveillance pricing possible – at that point, it's practically inevitable. This is what it means to create an enshittogenic policy environment: by arranging policy so that the most awful schemes of the worst people are the most profitable, you guarantee that those people will end up organizing commercial and labor markets.
When surveillance pricing is applied to labor, we call it "algorithmic wage discrimination," a term coined by Veena Dubal based on her research with Uber drivers:
Uber uses historic data on drivers to make inferences about how economically precarious they are, and then extracts a "desperation premium" from their wages. Drivers who are pickier about which rides they accept ("pickers") are offered higher wages than drivers who take any ride ("ants"):
On the back-end, Uber is inferring that the reason an ant will accept a worse job is that they have fewer choices – they are more strapped for cash and/or have fewer options for earning a higher wage.
This is a straightforward form of algorithmic wage discrimination, using the blunt signal of how discriminating a driver is when signing onto a job to titer the subsequent wage offered to that driver. More sophisticated forms of algorithmic wage discrimination draw on external sources of data to set the price of your labor.
That's the situation for contract nurses, whose traditional brick-and-mortar staffing agencies have been replaced by nationwide apps that market themselves as "Uber for nursing." These apps use commercial surveillance data from the unregulated data-broker sector to check on how much credit card debt a nurse is carrying and whether that debt is delinquent to set a wage: the more debt you have and the more dire your indebtedness is, the lower the wage you are offered (and therefore the more debt you accumulate – lather, rinse, repeat):
Surveillance wages are now proliferating to other parts of the economy, as "consultancies" offer software to employers that let them set all parts of your compensation – base wage, annual raises, and bonuses – based on your perceived desperation, as derived from commercial surveillance data that has been collected about you:
Genna Contino's Marketwatch article on the phenomenon offers a concise definition of "surveillance wages":
a system in which wages are based not on an employee’s performance or seniority, but on formulas that use their personal data, often collected without employees’ knowledge.
This means that carrying a credit-card balance, taking out a payday loan, or even discussing your indebtedness on social media can all lead to lower wages in the future. Contino references a recent report released by Dubal and tech strategist Wilneida Negrón, surveying 500 large firms, which concluded that surveillance wages are now being offered in sectors as diverse as "healthcare, customer service, logistics and retail." Customers for surveillance wage tools include "Intuit, Salesforce, Colgate-Palmolive, Amwell and Healthcare Services Group":
After a brief crackdown under Biden, the Trump regime has been extraordinarily welcoming to surveillance pricing companies, dropping investigations and cases against firms that engaged in the practice. A few states are stepping in to fill the gap, with New York state passing a rule requiring disclosure of surveillance pricing – a modest step that was nevertheless fought tooth-and-nail by the state's businesses.
In Colorado, a new House bill called the "Prohibit Surveillance Data to Set Prices and Wages Act" would prohibit the use of personal information in wage-setting:
https://leg.colorado.gov/bills/hb25-1264
This bill hasn't passed yet, but it's already doing useful work. Companies universally deny using surveillance data to set wages, insisting that they merely pay for consulting services that give them advice on how they could do surveillance wages – but don't actually take that advice. However, these same companies – including Uber and Lyft – are ferociously lobbying against the bill, raising an obvious question, articulated by the bill's co-sponsor Rep Javier Mabrey (D-1): if these companies don't pay surveillance wages, then "what is the problem of codifying in law that you’re not allowed to?"
Surveillance wages are a rare profitable use-case for AI, in part because surveillance wages don't need to be "correct" in order to be effective. An employee who is offered a wage that's slightly higher than the lowest sum they'd accept still represents a savings to the company's wage-bill. As ever, AI is great for fully automating tasks if you don't care whether they're done well:
The fact that surveillance wages are calculated by external contractors enables employers to engage in otherwise illegal price-fixing. If all the garages in town set mechanics' wages using the same surveillance pricing tool, then a mechanic looking for a job will get the same lowball offer from all nearby employers. If those bosses were to gather around a table and fix the wage for any (or all) mechanics, that would be wildly illegal, but the fact that this is done via a software package lets the bosses claim they're not actually colluding.
This is a common practice in other forms of price-fixing. We see it in meat, potato products, and, of course, rental accommodations (hey there, Realpage!). It's a genuinely stupid ruse based on the absurd idea that "it's not a crime if we do it with an app":
Speaking of crimes that are implausibly deniable when undertaken with an app: surveillance wages also allow employers to offer lower wages to women and brown and Black people while maintaining the pretense that they're in compliance with laws banning gender and racial discrimination.
In the wider economy, women and racialized people are already offered lower wages and – thanks to the legacy of racial discrimination in employment and housing – are more likely to be indebted:
By tapping into data brokers' dossiers that reveal the economic precarity of jobseekers, surveillance pricing allows employers to systematically lower the wages of women and Black and brown people, who have the highest incidence of indebtedness, while still claiming to offer race- and gender-blind wages. This is a phenomenon that Patrick Ball calls "empiricism washing": first, move the illegal racist discrimination into an algorithm, then insist that "numbers can't be racist."
But this isn't just about lowering wages at the bottom of the employment market. In recent history, the employers most eager to illegally lower their workers' wages are tech bosses, who had to pay massive fines for illegally colluding on "no poach" agreements to suppress the earning power of high-paid computer programmers:
(This is why the tech industry is so horny for AI – tech bosses can't wait to fire a ton of programmers and use the resulting terror to force down the wages of the remaining tech workers:)
Which means that the very programmers who write and maintain the surveillance wage software used on the rest of us are especially likely to have the tools they created turned on them.
Instacart reaches into your pocket and lops a third off your dollars
I'm at the end of my tour for my new book, the international bestseller Enshittification. My last two stops are CCC in Hamburg, Dec 27-30 and the Tattered Cover in Denver (Jan 22). Hope to see you!
There's a whole greedflation-denial cottage industry that insists that rising prices are either the result of unknowable, untameable and mysterious economic forces, or they're the result of workers having too much money and too many jobs.
The one thing we're absolutely not allowed to talk about is the fact that CEOs keep going on earnings calls to announce that they are hiking prices way ahead of any increase in their costs, and blaming inflation:
Nor are we supposed to notice the "price consultancies" that let the dominant firms in many sectors – from potatoes to meat to rental housing – fix prices in illegal collusive arrangements that are figleafed by the tissue-thin excuse that "if you use an app to fix prices, it's not a crime":
And we're especially not supposed to notice the proliferation of "personalized pricing" businesses that use surveillance data to figure out how desperate you are and charge you a premium based on that desperation:
Surveillance pricing – when you are charged more for the same goods than someone else, based on surveillance data about the urgency of your need and the cash in your bank account – is a way for companies to reach into your pocket and devalue the dollars in your wallet. After all, if you pay $2 for something that I pay $1 for, that's just the company saying that your dollars are only worth half as much as mine:
The economy is riddled with surveillance pricing gouging. You are almost certainly paying more than your neighbors for various items, based on algorithmic price-setting, every day. Case in point: More Perfect Union and Groundwork Collaborative teamed up with Consumer Reports to recruit 437 volunteers from across America to login to Instacart at the same time and buy the same items from 15 stores, and found evidence of surveillance pricing at Albertsons, Costco, Kroger, and Sprouts Farmers Market:
The price-swings are wild. Some test subjects are being charged 23% more than others. The average variance for "the exact same items, from the exact same locations, at the exact same time" comes out to 7%, or "$1,200 per year for groceries" for a family of four.
The process by which your greedflation premium is assigned is opaque. The researchers found that Instacart shoppers ordering from Target clustered into seven groups, but it's not clear how Instacart decides how much extra to charge any given shopper.
Instacart – who acquired Eversight, a surveillance pricing company, in 2022 – blamed the merchants (who, in turn, blamed Instacart). Instacart also claimed that they didn't use surveillance data to price goods, but hedged, admitting that the consumer packaged goods duopoly of Unilever and Procter & Gamble do use surveillance data in connection with their pricing strategies.
Finally, Instacart claimed that this was all an "experiment" to "learn what matters most to consumers and how to keep essential items affordable." In other words, they were secretly charging you more (for things like eggs and bread) because somehow that lets them "keep essential items affordable."
Instacart said their goal was to help "retail partners understand consumer preferences and identify categories where they should invest in lower prices."
Anyone who's done online analytics can easily pierce this obfuscation, but for those of you who haven't had the misfortune of directing an iterated, A/B tested optimization effort, I'll unpack this statement.
Say you have a pool of users and a bunch of variations on a headline. You randomly assign different variants to different users and measure clickthroughs. Then you check to see which variants performed best, and dig into the data you have on those users to see if there are any correlations that tie together users who liked a given approach.
This might let you discover that, say, women over 40 click more often on headlines that mention kittens. Then you generate more variations based on these conclusions – different ways of mentioning kittens – and see which of these variations perform best, and whether the targeted group of users split into smaller subgroups (women over 40 in the midwest prefer "tabby kitten" while their southern sisters prefer "kitten" without a mention of breed).
By repeatedly iterating over these steps, you can come up with many highly refined variants, and you can use surveillance data to target them to ever narrower, more optimized slices of your user-base.
Obviously, this is very labor intensive. You have to do a lot of tedious analysis, and generate a lot of variants. This is one of the reasons that slopvertising is so exciting to the worst people on earth: they imagine that they can use AI to create a self-licking ice-cream cone, performing the analysis and generating endless new variations, all untouched by human hands.
But when it comes to prices, it's much easier to produce variants – all you're doing is adding or subtracting from the price you show to shoppers. You don't need to get the writing team together to come up with new ways of mentioning kittens in a headline – you can just raise the price from $6.23 to $6.45 and see if midwestern women over 40 balk or add the item to their shopping baskets.
And here's the kicker: you don't need to select by gender, racial or economic criteria to end up with a super-racist and exploitative arrangement. That's because race, gender and socioeconomic status have broad correlates that are easily discoverable through automated means.
For example, thanks to generations of redlining, discriminatory housing policy, wage discrimination and environmental racism, the poorest, sickest neighborhoods in the country are also the most racialized and are also most likely to be "food deserts" where you can't just go to the grocery store and shop for your family.
What's more, the private equity-backed dollar store duopoly have waged a decades-long war on community grocery stores, enveloping them with dollar stores that use their access to preferential discounts (from companies like Unilever and Procter & Gamble, another duopoly) to force grocers out of business:
Then these dollar stores run a greedflation scam that is so primitive, it's almost laughable: they just charge customers much higher amounts than the prices shown on the shelves and price-tags:
When you live in a food desert where your only store is a Dollar General that defrauds you at the cash-register, you are more likely to accept a higher price from Instacart, because you have fewer choices than someone in a middle-class neighborhood with two or three competing grocers. And the people who live in those food deserts are more likely to be poor, which, in America, is an excellent predictor of whether they are Black or brown.
Which is to say, without ever saying, "Charge Black people more for groceries," Instacart can easily A/B split its way into a system where they predictably and reliably charges Black people more for groceries. That's the old cod-Marxism at work: "from each according to their desperation."
This is so well-understood that anyone who sets one of these systems in motion should be understood to be deliberately seeking to do racist profiteering under cover of an algorithm. It's empiricism-washing: "I'm not racist, I just did some math" (that produced a predictably racist outcome):
This is the dark side and true meaning of "business optimization." The optimal business pays its suppliers and workers nothing, and charges its customers everything it can. Obviously, businesses need to settle for suboptimal outcomes, because workers won't show up if they don't get paid, and customers won't buy things that cost everything they have⹋.
⹋ Unless, of course, you are an academic publisher, in which case this is just how you do business.
A business "optimizes" its workforce by finding ways to get them to accept lower wages. For example, they can bind their workers with noncompete "agreements" that ban Wendy's cashiers from quitting their job and making $0.25 more per hour at the McDonald's next door (one in 18 American workers have been locked into one of these contracts):
Or they can lock their workers in with "training repayment agreement provisions" (TRAPs) – contractual clauses that force workers to pay their bosses thousands of dollars if they quit or get fired:
But the most insidious form of worker optimization is "algorithmic wage discrimination." That's when a company uses surveillance data to lower the wages of workers. For example, contract nurses are paid less if the app that hires them discovers (through the unregulated data-broker sector) that they have a lot of credit-card debt. After all, nurses who are heavily indebted can't afford to be choosy and turn down lowball offers:
This is the other form of surveillance pricing: pricing labor based on surveillance data. It's more cod-Marxism: "From each according to their desperation."
Forget "becoming ungovernable": to defeat these evil fuckers, we have to become unoptimizable:
How do we do that? Well, nearly every form of "optimization" begins with surveillance. They can't figure out whether they can charge you more if they can't spy on you. They can't figure out whether they can pay you less if they can't spy on you, either.
And the reason they can spy on you is because we let them. The last consumer privacy law to pass out of Congress was a 1988 bill that bans video-store clerks from disclosing your VHS rental history. Every other form of consumer surveillance is permitted under US federal law.
So step one of this process is to ban commercial surveillance. Banning algorithmic price discrimination is all well and good, but it is, ultimately, a form of redistribution. We're trying to make the companies share some of the excess they extract from our surveillance data. But predistribution – ending surveillance itself, in this case – is always far more effective than redistribution:
How do we do that? Well, we need to build a coalition. At the Electronic Frontier Foundation, we call this "privacy first": you can't solve all the internet's problems by fixing privacy, but you won't fix most of them unless we get privacy right, and so the (potential) coalition for a strong privacy regime is large and powerful:
But of course, "privacy first," doesn't mean "just privacy." We also need tools that target algorithmic pricing per se. In New York State, there's a new law that requires disclosure of algorithmic pricing, in the form of a prominent notification reading, "THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA."
This is extremely weaksauce, and might even be worse than nothing. In California we have Prop 65, a rule that requires businesses to post signs and add labels any time they expose you to chemicals "known to the state of California to cause cancer." This caveat emptor approach (warn people, let them vote with their wallets) has led to every corner of California's built environment to be festooned with these warnings. Today, Californians just ignore these warnings, the same way that web users ignore the "privacy policy" disclosures on the sites they visit:
The right approach isn't to (merely) warn people about carcinogens (or privacy risks). The right approach is regulating harmful business practices, whether those practices give you a tumor or pick your pocket.
Under Biden, former FTC chair Lina Khan undertook proceedings to ban algorithmic pricing altogether. Trump's FTC killed that, along with all the other quality-of-life enhancing measures the FTC had in train (Trump's FTC chair replaced these with a program to root out "wokeness" in the agency).
Today, Khan is co-chair of Zohran Mamdani's transition team, and she will use the mayor's authority (under the New York City Consumer Protection Law of 1969, which addresses "unconscionable" commercial practices) to ban algorithmic pricing in NYC:
Khan wasn't Biden's only de-optimizer. Under chair Rohit Chopra, Biden's Consumer Finance Protection Bureau actually banned the data-brokers who power surveillance pricing:
These are efforts to optimize corporations for human thriving, by making them charge us less and pay us more. For while we are best off when we are unoptimizable, we are also best off when corporations are totally optimized – for our benefit.
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
