Last year, I wrote an explainer on algorithmic pricing that defined personalized pricing in terms of targeted discounts to allow consumers to participate in markets that they would be priced out of under a single price. A central takeaway was that when personalized discounts create additional mutually beneficial transactions that would not occur at the uniform price, those transactions increase economic surplus.
In a recent piece in The Sling, Gavin Sicard and Hal Singer take aim at my case that personalized pricing can expand access to goods. Unfortunately, their argument implicitly relies on two enormous and implausible assumptions, namely that a seller possesses both unrestrained monopoly power and the ability to read each consumer’s mind and determine their exact maximum willingness to pay for a given good.
Here is the story they tell:
“For instance, with personalized pricing, our example grocer can profitably lower prices from $3.00 down to $2.50 for consumers who value eggs less than their current market price. Yet these customers’ newfound ability to consume does not imply they are better off. Again, personalized pricing sets prices at a consumer’s maximum willingness to pay. In the jargon of economics, if a consumer is willing to pay $2.65 for a dozen eggs and is charged $2.65, they are indifferent between having the dozen eggs or having the cash to buy something else: their economic position remains unchanged. […] This nascent threat must be stopped before it pervades the economy[.]”
In other words, they argue that a hypothetical store with an implausibly large profit margin and a mind-reading algorithm takes every penny from every shopper, so the law should ban the discounts real stores offer real people.
There are a few problems with this analysis, most prominently:
- First, real personalized pricing algorithms make educated guesses and cannot read consumers’ minds.
- Second, real grocers operate on famously thin net margins and face vigorous competition; personalized pricing among competing sellers can therefore turn into a battle to offer targeted discounts and can lower prices for many consumers. More price-sensitive consumers, who are often lower-income, are particularly likely to benefit.
Personalized Pricing Is an Educated Guess, Not Mind Reading
Sicard and Singer breezily equate personalized pricing with sellers having perfect information about consumers’ willingness to pay, but that assumption is at odds with reality. The store doesn’t observe what you’d pay. All the store observes is a noisy estimate built from a finite amount of imperfect data. The store also faces a brutally asymmetric bet due to the existence of competitors: if the store guesses a little low, it gives up a few cents of margin; if it guesses a little high, it loses the entire sale.
Many stores facing that tradeoff under uncertainty deliberately price below their best guess, leaving a cushion. In the real world, personalized pricing is not the same as charging every consumer their precise willingness to pay. It looks much more like the store offering discounts to consumers that have shown price sensitivity to avoid losing their business. As price sensitive customers are often lower income, this can enhance both equity and social welfare, accounting for the higher utility value of savings to consumers who have tighter budgets.
How imperfect is the store’s data in practice? In a leading field study, published in the Journal of Political Economy, Jean-Pierre Dubé and Sanjog Misra ran a randomized pricing experiment with a real company, using machine learning on a large set of customer variables. Even then, the algorithm couldn’t come close to pinpointing what each customer would pay. More than 60% of customers would have paid less under personalized pricing than under a single profit-maximizing price applied to all consumers.
You Can’t Assume Away Competition
Sicard and Singer’s hypothetical grocery store breaks even at $2.50 and charges $3.00 in the single price world. That’s a 20% markup, held indefinitely, with no rival undercutting it.
American grocery stores look nothing like that. The Food Industry Association puts average net profit for food retailers at about 2%. If grocers could extract everything shoppers were willing to pay, it would show up in their financial statements. Spoiler alert: It doesn’t. Grocery is a sector with famously thin margins.
Another feature they gloss over is that the maximum price a consumer is willing to pay is not the same as the price a retailer can actually get away with charging a consumer, and that more important latter price is not a fixed number waiting to be identified. It varies over time and circumstance, and is capped by the price at the store down the road (plus the cost of traveling there). With real substitutes, personalized pricing can’t extract more than your switching cost, and for eggs bought weekly, a short trip from a competitor, that number is small. In the modern era with easy real-time price comparisons available to every consumer on their phone, competition rigorously disciplines prices.
When Sellers Compete to Personalize Prices, Competition Often Pushes Prices Down
The Sicard and Singer article never addresses what happens when competing stores can both personalize prices, because they assume away the existence of competitors. But in the real world, each competitor aims its sharpest discounts at exactly the customers the other most wants to keep, and the economic literature has reached a set of standard results showing us what to expect.
Thisse and Vives (1988) showed in the American Economic Review that competing firms get dragged into targeted pricing even though it can leave both with lower profits than simple uniform pricing. Kenneth Corts (1998) showed in the RAND Journal of Economics that targeting can even trigger “all-out competition” that lowers prices for every customer group. Shaffer and Zhang (1995) similarly analyzed targeted coupons, finding that when rivals both aim discounts at each other’s customers, the firms can find themselves in a prisoner’s dilemma, with shoppers benefitting from larger discounts. That helps explain why some firms publicly commit to everyday-low-pricing and no-haggle policies.
Rhodes and Zhou (2024), in the American Economic Review, looked at markets where nearly everyone buys, which is apt as nearly everyone buys groceries. They find that personalized pricing by competing firms can harm firms and benefit consumers. In the very setting Sicard and Singer chose, frontier theory predicts the opposite of their conclusion.
The FTC’s own spotlight shows concrete examples of the practice on the ground: targeted promotions and discount codes routed to specific shoppers, often infrequent or price-sensitive shoppers. In other words, they target digital coupons toward consumers apt to be swayed by the coupons, which are often lower-income consumers. A law against all personalized pricing would be, in large part, a law against discounts, and the shoppers who would most lose out are the price sensitive and lower-income consumers who were priced out of the market at the uniform price.
