Your first week with Lyrion
This is a day-by-day plan for your first five working days with Lyrion, written from what actually happens rather than what a tidy onboarding doc wishes happened. The theme is the same all week: go slow on purpose, and let the log earn the trust.
You will not save much time in week one. You are building the evidence that lets you save it in week two.
Day one and two: connect and watch
Day one is connecting the one queue you chose and the tools it touches, with the narrowest access that works, and approvals fully on. Do not automate anything yet. Let the agent draft, and read every draft it produces alongside how your team would have answered.
Day two is the same, and it should feel almost boring. You are calibrating. You want to see the agent handle a normal day's volume while a person still signs off on everything, so the first real evidence is honest.
Day three: read the log properly
By day three you have enough drafts to see patterns. Sit with the log, not the dashboard. Where is the agent consistently right. Where is it consistently wrong, and is the wrongness a missing document, a misread policy, or a case that should never have been in this queue.
Fix the fixable. Add the document it could not see. Tighten the rule it misread. Pull out the case type that does not belong. Most week-one problems are configuration, not capability, and the log tells you which.
Day four: take approvals off the safe slice
On day four, pick the safest, most mechanical slice of the queue and let the agent send those automatically, keeping approvals on everything else. This is the first real time saving, and it should feel small and undramatic. Small and undramatic is correct.
Watch that slice closely for the rest of the day. You are confirming that automatic behaves exactly like the approved behaviour did, because it should, since nothing changed except who clicks send.
Day five: decide the next step
Day five is a review, not a launch. Look at the week's log. Is approval starting to feel like a rubber stamp on more of the queue. If so, widen the automatic slice. If not, the log is pointing at the work still to do, and that is fine.
End the week with a queue that partly runs itself, a log you have read every day and trust, and a clear sense of the next slice or the next queue. That is a successful first week. Anything faster is usually a first week you redo later.
Why the record matters
A confident wrong answer is only dangerous if you cannot see how it happened. Every action here is logged, attributable and reversible, so a mistake is something you can find, explain and undo, rather than something you hear about from a customer weeks later.