User feedback

This sensor reads what real users say and do (feedback, support tickets, experiment results) in aggregate, with counts. It is the final judge of whether the product is wanted, and the slowest signal of all: it arrives over weeks.

What it sees

Whether a change helps the people it was made for: what they complain about, what they ask for, what they use and what they leave alone. No program or model can answer that on their behalf. It costs time, since a trend takes weeks of reports, and attention, since the items must be read and grouped. It hears only from users who write in or whose behaviour is measured, and those who write in are not a fair sample of all users.

Reading the signal

• A topic with its count: how many users raised it, and where. The count, not the tone, sets its weight.
• A single loud report: not a trend. It becomes one only if the count grows.
• An experiment result: whether users valued the change, judged by what they did rather than what they said.

A drop in reports can mean the channel broke (a form that fails, a support queue that moved) rather than that users are happier. Check that the channel still delivers before reading silence as satisfaction.