Run · Agent outcomes
"Is this agent working?" gets an actual answer.
Most companies find out an agent wasn't working at renewal time, from a gut feeling. Agent outcomes turns that gut feeling into a number, measured against the goals you actually set, updated continuously.
Why a generic score wouldn't work
"Working" means something different for every agent.
A support agent's job is resolving a question without escalation. A research agent's job is finding the right document. Scoring both against the same generic metric would tell you nothing useful about either. So the goals an outcome score is weighed against are set per agent, by you.
What makes up the score
Not a black-box number.
- Weighted health score
- Measured against goals you set for that specific agent, not a generic platform-wide number.
- Thumbs-up ratio
- 75% or higher reads green, 50% or higher reads amber, below that reads red, from reviewed answers, not guesses.
- Your goals, not ours
- A support agent and a research agent are judged against different things, because they should be.
- Visible on the tile
- The score sits on the same card as everything else in the Agent fleet. No separate report to pull.
Reviewed answers usually come from human-in-the-loop review; the score itself lives on every card in the Agent fleet.
FAQ
Companies reviewing agent performance are asking
Who sets the goals an outcome score is measured against?
You do, per agent. The health score is weighted against goals your team defines for that specific role, not a generic platform-wide metric.
What is the thumbs-up ratio, exactly?
The share of an agent's reviewed answers marked good rather than corrected. 75% or higher reads green, 50% or higher reads amber, and below that reads red, visible right on the fleet tile.
Does a low outcome score do anything automatically?
It surfaces the agent for attention, on the fleet grid and in Analytics, rather than taking an automatic action. A person still decides what changes.
Know an agent isn't working before renewal decides it for you.
Twenty minutes with the founder. No slides. The real product.
