What it surfaces
- Notable conversations — strong buying/contact/sign-up intent, qualified leads, long conversations that landed, trust questions handled well. Persisted as a registry of golden examples.
- Actions claimed without a real tool call — when the bot says “I’ve forwarded your request” but no tool actually fired. A serious, otherwise-invisible defect: only reading the conversation against the tool history catches it.
- Tool failures — success/failure rate per tool, with detail.
- Unmet demands — what visitors asked for that you don’t yet offer, clustered by theme.
- Suggestions — 3–5 concrete improvements (knowledge to add, a CTA to insert, a tool that’s missing or broken).
How notability is judged
The analysis receives the same context your bot has — its persona, knowledge base, and the tools it actually exposes — then judges each conversation against your bot’s mandate, not a fixed checklist. “Notable” means different things for an e-commerce store, a clinic, or a directory — the digest adapts because it reads your bot’s own configuration.Frequency
Weekly, daily, or monthly — depending on your plan — plus on-demand runs.- Scheduled digests email your configured recipients.
- On-demand: trigger one anytime (subject to a monthly quota) from the API or MCP.