business · reviewed protocol

Run a controlled support-triage pilot

Compare a human-only baseline with an AI-assisted draft-and-review workflow without automating the final customer decision.

Fork into workbenchShare stable link
Version
1.0
Reviewed
2026-08-24
Next review
2026-11-22
Reviewer
AI Amigos Editorial Desk
Scope
Protocol structure, privacy boundary, and measurement method

Job to be done

Test whether AI can reduce triage time while preserving routing accuracy and human approval.

Evidence state: Editorial reference protocol; no community outcome is claimed.

Baseline

  1. Measure median triage minutes per case
  2. Record routing accuracy on a fixed reviewed sample
  3. Record escalation and rework rates

Method

  1. Select 30 non-sensitive historical cases
  2. Blind the reviewer to the workflow variant
  3. Run human-only and AI-assisted variants
  4. Require human approval before any external action

Evaluation

  • Median minutes per case
  • Routing accuracy
  • Critical miss count
  • Reviewer override rate

Limitations

  • The protocol is an editorial starting point, not evidence that AI improves support work.
  • Do not include customer personal data unless an approved data-processing agreement and access control exist.