TR-AI

Staged releases

Staged releases

Bundle settings changes into a draft, test it with past tickets and simulated customers, roll it out to a few categories, then publish or roll back.

All plans (roll-out by ticket category: Growth Pro) Open in TR-AI5 min read

Changing a threshold or the brand voice on a busy help desk affects every customer at once. A release lets you prepare those changes as a draft, check what they do, and switch them on together — with a one-click way back.

A release can changeTested before publishing?
Auto-send and review thresholds, per-category thresholdsYes
Paid web search, live site search, LLM-confidence overrideYes
Brand voice (tone, greeting, reply length, sign-off)Yes
Turning escalation rules on or offNo — applied when you publish
Turning procedures on or offNo — applied when you publish
  1. 1
    Create a draft
    Only fill in what should change; blank fields keep today’s value. A draft never affects live answers.
  2. 2
    Test
    Replays your newest regression suite and runs every simulated-customer scenario with the draft settings. Nothing is sent. Results appear on the release card with links to the details.
  3. 3
    Roll out to a few categories (Growth Pro)
    Apply the engine settings only to tickets in the categories you pick — for example billing — and watch the results before widening. Only one release can roll out at a time.
  4. 4
    Publish
    Everything goes live for every ticket at once and a new settings version is recorded on the Regression tests page.
  5. 5
    Roll back if needed
    Restores exactly the values the release replaced. Roll back the newest published release first.
Note:Safety settings — assisted mode, the output guard, PII handling, competitor rules and URL allowlists — can never be part of a release. The auto-send threshold can never go below 82%.
Tips & tricks
  • Give releases names that say why (“Fewer billing escalations”) — the history reads like a changelog.
  • Build your regression suite and a handful of scenarios first; a release test is only as good as what it replays.
  • Roll threshold changes out to one category for a few days before publishing to everyone.
  • If a published release surprises you, roll it back first and investigate afterwards — it takes seconds.
PreviousSimulated customersNextSimulation Lab and backtesting