Escalation Rate

The proportion of agentic task executions that require escalation to a human Steward — the primary operational metric governing the boundary between autonomous execution and human judgment under the Stewardship Model.

The Escalation Rate is distinct from MTTI. MTTI measures the temporal dimension of autonomous operation — how long the system runs between required human interventions. The Escalation Rate measures the transactional dimension — what proportion of individual task executions require a human decision before they can complete. The two metrics together define the operational profile of an autonomous system: MTTI tells you how frequently the system calls for help; Escalation Rate tells you how often, across all the work it handles, a human is in the loop.

The Escalation Rate is not a fixed target across the portfolio. It is tier-dependent, topic-dependent, and risk-dependent. At T1 — routine, scripted, high-volume tasks with binary outcomes and low stakes — Arco targets an Escalation Rate of approximately 1:100: one human intervention per hundred autonomous executions. The customer care simulation data supporting this target shows T1 ticket escalation rates of 1–5% across password resets, FAQ handling, order tracking, and basic billing — with the lowest-complexity tasks achieving 1% escalation, the operational expression of the 1:100 target. At T2, where tasks require moderate judgment and contextual interpretation, the rate rises to approximately 1:5 to 1:10. At T3, where regulatory compliance, relationship management, or high-stakes judgment is required, human involvement is mandatory at a proportion that reflects the specific risk profile of the task type.

First used: April 2026