Every conversation lands in a table as a summary

A month into running a bot a question appears that nothing can answer: what do customers actually ask, and where does the bot fall over? The conversations exist; nobody is going to read them. A summary per conversation turns them into data.

What is inside

1. A customer message — the entry.
2. Answer to the customer — an ordinary agent with conversation memory.
3. The whole conversation — a history node. Memory feeds the model; history hands over the conversation entire, and the summary needs the second.
4. Conversation summary — the model returns the topic, the outcome and the customer’s mood.
5. Row for the table and Store the summary.

The scenario writes about itself. Data usually comes from outside; here the source is its own conversation, and that is the only way to get this statistic without reading transcripts.

What to set up for yourself

• The summary fields. Topic, outcome, mood is the minimum to start with. “Did it need a human” and “was it in the documents” are the obvious next two.
• When to write. A row is written per message today, so the table accumulates one per turn. That is deliberate simplicity: writing once at the end needs a “the conversation ended” signal, and there is none — customers just leave.
• The collection on the write node.

What to connect

A collection under Tables. From there it answers questions like “how many enquiries were about delivery” the same way “How much did I spend” does.