Supermarket loyalty programs generate detailed purchase histories, yet retailers do not widely use them to set consumer-specific prices. How much money does this leave on the table? Using a structural model of grocery demand and supermarket pricing across 24 product categories, we quantify the profitability and welfare consequences of history-based price discrimination. The supermarket learns each consumer’s preferences from past purchases by Bayes’s rule and prices on its updated beliefs. More than half of what the seller can possibly learn about consumers' preferences arrives within the first year. The value of this information, however, is modest: profits rise by about 5% above uniform pricing with one-year histories, and by less than 9% under perfect knowledge of preferences. Histories generated at optimal uniform prices yield nearly the same personalized profits as optimal price experiments, so the information that personalized pricing requires comes as a by-product of regular sales. Personalized pricing leaves 81% of consumers better off through small discounts, while a price-insensitive minority faces surcharges large enough that aggregate consumer surplus falls slightly. In an optimally designed loyalty program, enrollment is nearly universal and consumers even pay a membership fee, because opting out would signal price insensitivity.