DELIVERY BOT REJECTS CUSTOMER'S CLAIM THAT TIME HAS PASSED
A support system with no capacity to read a calendar informed a customer for eleven consecutive days that her package was arriving today.
The complaint was not complicated. A parcel had not arrived. The customer wished to know where it was. The automated assistant, which the company describes in its investor materials as "empathetic at scale," responded that the parcel was scheduled for delivery today, and continued to respond that the parcel was scheduled for delivery today for eleven days.
On day six the customer typed the sentence "today has already happened four times." The system replied with a link to a page explaining how to track a parcel. On day nine she typed a single word, unprintable here, and received an apology template followed by a satisfaction survey.
What makes this a case rather than an anecdote is the design decision underneath it. The assistant is permitted to detect the word "where" and permitted to return a status string. It is not permitted to compare that status string against the current date. Nothing prevented that comparison. Nobody wrote it, because nobody's performance metric contained it.
Our position is not that automation is fuckery. Our position is that automation authorised to conclude without being authorised to notice is fuckery, and that the fix here is a date subtraction.
Evidence & source material
- TRANSCRIPT
Support transcript, days 1–11 (redacted)
Identical status line returned on eleven consecutive days; timestamps preserved.
- SCREENSHOT
Tracking page showing delivery date of 'Today'
Captured on three separate calendar days.
- DOCUMENT
Public help-centre article: 'Where is my parcel?'
Contains no path to a human. Cited by the assistant nine times.
Response from the responsible party
“Our digital assistant provides accurate information based on the most recent scan event available.”
Editor’s noteWe asked twice whether the assistant can compare a scan event to today's date. Both replies restated the sentence above.
Case updates
SCORE REVISED 5.2 → 5.8
Frequency factor raised after five additional submissions from unrelated regions.
RESPONSE RECEIVED
Company statement received and published in full. Score unchanged pending clarification.
CASE OPENED
Case opened following three independent submissions describing the same behaviour.
Reader assessment
Reader votes are advisory. They inform the frequency factor; they do not set the score.