Capturing Rovo Agent conversations for review and continuous improvement

[Marc Kelvin O_ Tan]

We've built a Rovo Agent as a customer-facing self-service assistant (help center + Confluence pages mapped to intent/request types) to reduce inbound tickets to our service desk. Our go-live bar was deliberately low, it doesn't need to be perfect on day one, but we do need to review past conversations afterward to spot knowledge gaps and improve our docs.

The gap: native conversation history/analytics export for Rovo Agents isn't available yet, so there's no built-in way to pull "what are people asking and where are we falling short" directly from the platform.

What we've done so far: we tried working around it by having the agent write its Q&A exchanges to a Confluence page for later review (a pattern we heard about from another local user group). Two limitations showed up in our implementation:

1. It only fires at the end of the conversation — no per-turn or mid-session capture, so a user who drops off without a clean close leaves nothing logged.
2. It requires the user to explicitly action a "publish" step through the chatbot — the agent asks the user to publish the article themselves, which pulls them out of the self-service flow at the exact point they need a fast answer.

What we're looking for: any references or approaches for taking this further —

1. Getting conversation capture to Confluence (or elsewhere) working without a user-triggered publish step — service-side action, webhook, scheduled export, anything along those lines.
2. Capturing mid-session state, not just end-of-conversation.
3. Any other patterns people are using to get conversation-level visibility out of a Rovo Agent today.
4. Whether native chat history/review for Rovo Agents is on a visible roadmap.

Appreciate any pointers, happy to share back what we land on.