Every AI Agent conversation is logged in your Unified Inbox alongside conversations handled by your team. There is no separate 'AI Agent inbox'.
Where to find them
In the Unified Inbox, AI Agent conversations appear in the same view as human-handled ones. They are tagged so you can filter to AI Agent-only conversations when needed.
What to look at in the first weeks
After going live with a new AI Agent, review conversations regularly. Three things to check:
1. Where did the AI Agent hand off? Look at conversations marked Bot Left. If the AI Agent is leaving questions it should be able to answer, the knowledge base has a gap. Filter the agent's Knowledge tab by the relevant tag to check whether something on that topic already exists and is simply worded badly — and if nothing exists, add it. Writing the item by hand is one route; importing from your website or a policy document is often faster. See Importing knowledge into your AI Agent.
2. What is the AI Agent answering correctly? Spot-check resolved conversations. The AI Agent will confidently quote whatever you wrote, so the test is whether the answer was actually correct and complete.
3. Are there patterns in the questions guests ask? Often, the AI Agent reveals questions you did not realise guests had. Add knowledge items to cover them, and tag them as you go so the next review is quicker.
Consider starting in draft mode
If you would rather review the agent's answers before a guest sees them rather than after, switch the agent into draft mode for the first few weeks. Its replies then wait in the operator's composer to be sent, edited or rejected, so nothing reaches a guest unchecked.
It also makes this review loop faster: an answer your operators keep editing the same way is a knowledge item that needs fixing, and you find that out immediately rather than by reading transcripts later. See Draft mode: letting an operator check the agent before it sends.
Using the Inbox Dashboard
If you have Bookboost Insights, the Inbox Dashboard surfaces useful patterns:
An AI summary clustering frequent conversation topics, plus an average sentiment score. A new pattern in this view is usually the next knowledge item to write.
Response times by channel and operator. How the AI Agent is doing against your standards.
Resolution time by operator and channel. How long AI Agent-handled conversations take to close compared with human-handled ones.
For more, see the Bookboost Insights documentation.
Keeping the review loop manageable
Tag knowledge items as you create them. When a review turns up a problem with, say, arrival information, you can filter the Knowledge tab to that tag and read everything the agent holds on the subject in one go, instead of scrolling the whole list. The Untagged filter is worth a look too — it shows the items nobody has classified yet, which after a bulk import is usually most of them. See Organising knowledge with tags.
What 'good' looks like
Resolution rate climbing week over week as the knowledge base improves.
Handovers clustered around legitimately complex topics, not routine gaps the agent should have covered.
Operator feedback that handovers arrive with useful context, not garbled threads.
If you are running in draft mode: most drafts going out unedited.
Need Help?
Please contact us through the 'Talk to Us' option on the left menu in the platform, or through the Bookboost Support email at support@bookboost.io if you have questions or need additional support.