A foreign commuter cleaning out a second‑hand smart shelf he had bought online found a folder of pricing logs that, when examined, detailed how an AI system recommended different prices for the same packet of rice depending on who stood in front of the shelf. The finder, Scot O'Connell, brought the files to a friend with data skills; within days the trail had reached a major Tokyo start‑up and national retail chains.
The logs carry timestamps, store IDs and what appear to be hashed device identifiers next to percentage mark‑ups. The folder bore the internal project name 'ApertureDynamic' and references to a Tokyo company called ApertureAI Japan, a fast‑growing supplier of machine‑learning merchandising tools used by convenience stores and supermarket groups across Japan.
How the files surfaced and what they show
Forensic examination by independent analyst Jenna Ramcharan shows the logs pair short‑term wireless probes — small pings from phones and loyalty cards — with a customer scoring model and an 'optimal price' output. "The system was not simply forecasting demand; it was tailoring individual price suggestions," Ramcharan said. Her analysis, shared with this paper, includes anonymised examples where identical customers in the same hour were recommended discounts or surcharges of up to 18 per cent on essential items.
They were literally pricing my shopping to my phone.
Scot O'Connell, commuter who found the files
Company and retailers respond
ApertureAI Japan, whose valuation topped ¥120 billion (about US$800m) in a 2025 funding round, issued a written statement saying its software is designed to 'optimise shelf space and reduce waste' and that clients set final prices. Masato Fujikura, a company spokesperson, told The Plausible Post: "Our tool provides demand projections and promotional suggestions. No device transmits personal identifiers to us; any hashes found are hashed by clients in aggregate." He declined to answer whether the company had contracts that specified individualised price outputs.
Several retail chains approached by the paper said they were examining the logs. A mid‑level procurement manager at a Tokyo convenience chain, speaking on condition of anonymity because they were not authorised to discuss the matter, said the company had installed ApertureAI modules in more than 500 outlets and that staff were 'shocked' by the suggestion of per‑customer mark‑ups. Japan's Consumer Affairs Agency confirmed it had received informal complaints and was coordinating with the Competition Policy Bureau to determine whether rules on unfair pricing were breached.
Market reaction was swift. Shares in listed retailers that had disclosed partnerships with ApertureAI Japan fell an aggregate 6.3 per cent the day after the files became public; technology investors pulled roughly ¥15 billion from venture funds with exposure to personalised retail software, according to market data. Consumer groups in Osaka and Tokyo have begun crowding stores to photograph digital shelf labels and raise refunds, and Diet members from several parties said they would demand hearings on algorithms that affect household bills.
What happens next is uncertain. Regulators can seek fines and require changes to disclosure if they find discriminatory pricing practices, but legal pathways are untested for algorithmic personalisation. For now, consumers are checking receipts and retailers are unplugging modules they say they installed to improve stock rotation. "If software can change what you pay depending on a phone sitting in your pocket, that's a public‑policy problem," said Esma Polat, an economist at Meiji Private Research, who is advising a group of consumer advocates reviewing the logs.
