signalisys
Readiness glossary

Human-in-the-Loop Design

How agents hand off to humans — confidence thresholds, approval flows, and the feedback loop from every escalation back into the agent.

Why this score moves

  • Existence of confidence scoring on agent actions
  • Whether escalation reasons feed back into design
  • Coverage of role-based intervention rights

Reading the four bands

EMERGING

No defined escalation; agents either over-act or stall — humans can't intervene cleanly.

DEVELOPING

Escalation exists but is binary (handoff or not) and doesn't capture the reason or fix.

ESTABLISHED

Tiered escalation with confidence thresholds and a feedback loop into prompt / data updates.

OPTIMISED

Closed-loop human review where every escalation tightens future agent performance.

What good looks like

  • Confidence thresholds tied to specific actions
  • Tiered escalation paths (assist → suggest → defer)
  • Captured reason codes for every human take-over
  • A feedback loop from escalations into prompts, actions and data
More dimensions
Data Cloud
Data Cloud Readiness
Agentforce
Agentforce Fit
Flow
Flow & Automation Depth
Core Cloud
Core Cloud Foundation
Trust
Trust & Governance
Exec
Executive Alignment