Escalation Agent

Software componentHuman OversightExperience & Human Oversightarc:EscalationAgent

A decision agent that identifies cases requiring human review using confidence, sensitive-content, authority-threshold and customer-history criteria, and publishes escalation events.

Responsibility. Decides which cases require human review.

Also known as: Escalation decision, Fallback hierarchy, Escalation triggers, Tiered oversight routing, Sentiment monitoring agent

Variant of Oversight Gate abstract

Direct neighbourhood (hover for relationship types)

Relationships

receives data from dynamic

sends data to dynamic

Event Broker abstract Ch1.3

triggers dynamic

Escalation Handler Ch10.1 Ch10.5

is orchestrated by control

Supervisor Agent Ch1.3

monitors assurance

Worker Agent abstract Ch10.5

Design guidance

SHOULD evaluate criteria beyond model confidence (sensitive content, authority limits, repeat escalations).
SHOULD publish escalation events rather than call each downstream consumer directly.
SHOULD escalate out-of-competence scenarios to specialist agents or human experts rather than produce plausible but incorrect answers.
SHOULD prioritise escalation of frustrated customers detected by sentiment analysis.
MUST honour explicit user requests for a human agent immediately.
SHOULD escalate on low confidence below defined thresholds and on repeated clarification failures.
MUST provide an easy escalation path to avoid trapped-user syndrome.
SHOULD escalate emotionally charged interactions regardless of confidence, and repeat contacts from the same customer within short timeframes.

Quantitative guidance

As stated by the sources; verify before use.

Refunds over $1000 require human approval (Ch1.3 worked example).
Klarna autonomously handles 65% of inquiries; Erica reduced call-center volume for routine inquiries by 50% (Ch10.1).

Classification

Patterns: Confidence-threshold escalationAuthority thresholdsFrustration detection from repeated concerns
Quality attributes: Safety (ISO/IEC 25010 | NIST AI RMF: safe)Transparency and accountability (NIST AI RMF: accountable and transparent)
Risks mitigated: Unreviewed high-impact actionsAmbiguous tickets handled autonomously

Sources

Ch1.3: T. Nguyen, "Multi-Agent Systems," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 1.3. ISBN: 9798244538229.
Ch3.3: T. Nguyen, "Web Navigation and Interaction Benchmarks," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 3.3. ISBN: 9798244538229.
Ch10.1: T. Nguyen, "Conversational UI," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 10.1. ISBN: 9798244538229.
Ch10.5: T. Nguyen, "Human-over-the-Loop," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 10.5. ISBN: 9798244538229.