How TR-AI works

What happens to a ticket from the moment it arrives until it is answered, reviewed or handed to your team.

All plans5 min read

TR-AI sits on top of your help desk. Every new ticket and every customer follow-up is read, matched against what your company already knows, answered if TR-AI is confident, and otherwise drafted for an agent or handed over with a summary. Your agents keep working in the help desk they already use.

Where answers come from (in this order)

1
Your knowledge base and past resolutions
Help-center articles, docs sites, Confluence and Notion pages, plus tickets your team resolved by hand. Searched by meaning and by exact keywords (product names, error codes).
2
A free search of your own website
If nothing above reaches the send threshold, TR-AI searches your public site live — useful for pages that are not in the knowledge base yet.
3
Paid web search — last resort
Only when nothing above reaches the send threshold (82% by default) and web search is enabled for your account. Results are limited to the domains you allow.
4
Procedures and live data
For “do something” tickets (order status, cancellations, refunds) a procedure you wrote can look up live data in Shopify, Stripe, Jira or an MCP server, with approvals for anything that changes data.

What happens next: send, review or triage

Confidence	Outcome	What the customer sees
82% or higher (default)	Sent automatically	The answer, in their language, as a public reply
60% – 82%	Review	Nothing yet — an agent approves, edits or rejects the draft
Below 60%, or a rule says so	Triage	An acknowledgement (optional); your team gets an internal note with a summary

Sensitive topics (billing disputes, legal, security), escalation rules you define, and anything TR-AI is unsure about are always handed to a person.

Tips & tricks

Start in assisted mode for a few days: every reply waits for approval, and you see exactly what TR-AI would send.
The ticket log on the Overview page (right below the numbers) shows every ticket with its score breakdown and the sources used — open one to see why a ticket went to review.
Most “TR-AI didn’t know” cases are knowledge-base gaps. The Topic insights and KB Health pages tell you which articles to write next.

PreviousConfidence scoring NextRolling out safely: assisted mode and the review queue