Collect feedback from every channel by topic
Feedback lands in a chat, an inbox and a marketplace, each watched by different people. The agent sorts all of it into one topic list and counts the shares.
Medium · 25 min · once a month
What you get
Feedback topics, April — 1,812 messages Topic Telegram Email Ozon Delivery 14% 9% 38% App behavior 31% 22% 6% Payment and refunds 12% 27% 9% Shares are counted inside each channel: people write more in chat, so 14% there and 38% on Ozon are not comparable. Unsorted: 96 messages, with the reason for each.
A sample on made-up data — your numbers will be your own.
Who it fits
- Three teams watch three different sources and nobody holds the whole picture.
- You need to tell a widespread topic from a noisy channel: chat produces more messages by nature.
When it won't work
There is no topic list and nobody to approve one — the split drifts from one run to the next.
How the agent does it
Approve the topic list
Give your own list or review the one the agent proposes — and look at that before the numbers. Reuse the same list next time or the runs will not compare.
Run the sorting
The agent assigns each message to one topic, sets the debatable ones aside and counts shares inside each channel separately.
Compare the channels
Look separately at topics with few messages and channels where the period is only partly covered. Those shares swing on a handful of messages.
What you set
Telegram, Email, Ozon. By hand: the period and the topic list — your own, or the one the agent proposes if you do not have one yet.
What you'll need
Starter prompt
Copy the prompt or open it straight in a chat with the agent.
Collect the feedback from Telegram, email and Ozon reviews for [period] and sort it by a single taxonomy [list of topics — or propose your own and show it for approval before the analysis]. One taxonomy for all three channels: the same topic is named identically in each. Assign every message to exactly one topic; put the debatable ones in "unclassified" and show them separately rather than splitting them across two topics at once. The channels differ in volume, so besides the count compute the topic's share within its own channel — compare shares, not counts. Do not add the channels together into one overall ranking. Output a "topic × channel" table: count, share within the channel, two quotes each — and a list of topics that exist in only one channel.
Open in chatCheck the classification. List: — topics you introduced beyond the approved taxonomy, and what landed in them; — messages that ended up in "unclassified", grouped by reason; — topics where a channel has fewer than [20] messages and the share jumps because of single items; — channels where the period is not fully covered — from which date to which there is data; — messages from the same person counted as several tickets; — which topics you can re-sort yourself and where a decision is needed on where something belongs.
A second prompt — the agent uses it to review its own work and show what's left for you.