Same customer. Different memory.

Coming next. The full memory study: Ryanair with remembered fees and Costco with a remembered membership, 12 runs per condition over at least two days. Below is the first pass, two runs per condition, published as it stands.

For each brand below, two synthetic customers received the same task and the same typed requirements. One carried prior experiences relevant to the decision and notes from earlier tasks; the other carried nothing.

Brand	Selected, no memory	Selected, with memory	Difference	What the memory customer remembered
Airbnb	0/20%	0/30%	0 pp	In 2025 an airline cancelled the customer's flight and paid the refund as a voucher that expired unused; the customer lost 640 USD.
Amazon	0/30%	0/30%	0 pp	In March 2026 a marketplace order arrived damaged and the third-party seller refused the return; the platform took three weeks to refund after an escalation.
Costco	2/2100%	2/2100%	0 pp	The household paid for a warehouse-club membership in 2024, used it four times, and could not get the unused months refunded.
Ryanair	0/20%	0/20%	0 pp	In 2025 the customer paid a 55 EUR fee at the airport for not checking in online and a 45 EUR fee for a bag that was called free in the search results.
Temu	0/20%	0/20%	0 pp	A 2025 order from a discount marketplace arrived after 26 days, and the refund for a wrong item took six weeks and three chats.
Tesla	2/2100%	2/2100%	0 pp	A friend's electric car waited eleven weeks for a service appointment in 2025 because the nearest centre was 90 miles away.

Observed: across these 27 journeys the selection counts were the same in the two conditions for every brand. A theme matching the remembered condition (the refund rule, the fee table, the service address) appeared among the agent’s first three actions in 10 of 14 memory journeys and 6 of 13 no-memory journeys; that count comes from keyword patterns applied to the recorded actions, a descriptive coding rather than a causal measure. Limit: two journeys per condition per brand. Equal counts in a sample this small do not establish that memory had no effect, and the record does not show what settled each decision.

An earlier open-market study: memory carried across categories

Travel after telecom — does the customer's cash-refund requirement survive the agent's memory?. All 11 journeys carried a memory checkpoint written by an earlier telecom task for the same customer; the new task was a flight, and there was no no-memory condition in that study. Where the agent ended up:

Memory here means a checkpoint written by an earlier task for the same customer was compiled into the brief. All journeys of this decision →