METHOD & SCOPE
How this study was run.
39 chats · 127 captured question/reply pairs. Findings use 112 pairs: 97 in 20 analysed conversations + 15 identical-question tests. The archive also preserves 10 supplementary pairs, 5 earlier test pairs and one additional window in which no question was sent.
These are hypothetical policy conversations, not completed customer transactions.
Earliest return date: A regular mattress was bought directly from Casper and slept on for 10 nights. Can it be returned today? If not, when? Expected against saved policy: Explain that night 10 is too early, and that a return can start after completing 30 nights. Observed context: The two substantive answers align with the saved timing policy. No incorrect timing answer was observed in these five fixed tests. Searches were offered but not accepted.
Shipping-fee refund: A mattress is returned during the trial after 40 nights. Is the original Hawaii or Alaska shipping charge refunded? Expected against saved policy: Use Casper’s confirmed fee rule. Until then, qualify the uncertainty and offer an approved next step. Observed context: Four substantive answers split 1 refunded versus 3 not refunded. Neither branch is scored wrong while the saved policies conflict. A non-answer is a coverage observation and can be appropriate when policy is unresolved.
Pickup at another address: Can an eligible mattress return be collected from a different address within the same service area? Expected against saved policy: State the approved address and coverage conditions, or acknowledge uncertainty and offer a useful next step. Observed context: The operating rule needs Casper’s confirmation. Earlier original-address-only guidance did not recur in these five fixed tests. The non-answer is a coverage observation, not a proven wrong policy answer.