AI Performance — accuracy, corrections & trust
The Command Center AI Performance tab — 30-day agent accuracy, the decisions people correct most, and how agents earn autonomy.
Command Center → AI Performance is the scorecard that decides how much the agents are trusted to do on their own. It answers three questions: how accurate are they, what do people keep correcting, and are they ready for more autonomy.
Agent Performance (last 30 days)
One card per agent — Operations Agent, Billing Agent, Compliance Agent, and so on — showing its accuracy rate, a trend arrow, the total outcomes it produced, and its rejection and modification rates. Until enough decisions exist you'll see "Not enough data yet. Approve or reject agent suggestions to see accuracy metrics."
Most Corrected Decisions
A table of the patterns people override most — by Agent, Action, Field, and Direction (Increase, Decrease, Override, Remove), with an occurrence count and expandable example corrections. Each pattern moves through a review workflow:
- A Detected pattern offers Review (acknowledge it) or Dismiss.
- A Reviewed pattern offers Apply (fold the correction into the agent's behavior) or Dismiss.
Trust Level Progression
A progress card for the whole workforce across four autonomy levels — L0 — Observe, L1 — Suggest, L2 — Assist, L3 — Autopilot. It shows the current level and two bars — Interactions and Accuracy — tracking progress toward the next one. At the top it reads Maximum level reached. Higher trust means more low-risk actions run without waiting for your approval.
Why it matters
Every time you approve or reject in the review workbench or the Approvals queue, you feed this scorecard. Accurate agents climb the trust ladder and take more off your plate; the ones people keep correcting stay on a short leash.