ICMD Agentic Feature Readiness Kit (2026)

Use this kit to turn an “AI feature” into a revenue-grade, auditable workflow.

1) Pick the workflow (one only)
- Name the workflow: ___________________________
- Trigger (what starts it?): _____________________
- Done state (what does “resolved” mean?): _______
- Frequency (weekly volume): ________
- Business owner (Ops/Finance/Support/Sec): ______

2) Define the autonomy ladder (ship in this order)
Tier A — SUGGEST: Agent drafts actions; human executes.
- Required UX: draft output, citations/inputs shown, copy/export.
Tier B — ASSIST: Agent executes tools only after approval.
- Required UX: action preview, approve/deny, partial execution, safe defaults.
Tier C — ACT: Agent auto-executes within policy.
- Required UX: policy rules, kill switch, rollback, post-run report.

3) Policy + permissions template
- Tools allowed (scoped): _______________________
- Data boundaries (what it must never access): ____
- Approval rules (examples):
  - If $ amount > ____ then require approval
  - If confidence < ____ then require approval
  - If record is enterprise/VIP then require approval
- Logging: retention ___ days; PII redaction yes/no; export format ______

4) Evaluation plan (minimum viable)
Offline eval:
- Collect 50–100 real historical tasks.
- Define “correct” as structured fields + validators.
Shadow mode:
- Run in production without execution for ___ days.
- Track: completion rate, top exception reasons, human edits.
Gated rollout:
- Start with Tier A for all users.
- Promote to Tier B for a pilot group (___ accounts).
- Promote to Tier C only when: policy violations < ___ per 1,000 runs AND rollback works.

5) Metrics and targets (CPRO-first)
Define CPRO (Cost per Resolved Outcome):
CPRO = (model cost + tool costs + human review time cost) / successful outcomes
- Baseline human cost per outcome: $____
- Target CPRO: $____ (must be clearly below baseline)
Operational metrics:
- Outcome completion rate target: ____%
- Human touches per outcome target: ____
- Policy violations target: ____ per 1,000
- p95 workflow latency target: ____ seconds

6) Incident and rollback runbook (must exist before Tier C)
- Global pause automation button: owner _______
- Per-tool disable (OAuth scopes / API keys): owner _______
- Rollback method (undo action / bulk revert script): __________
- Customer comms template: where published (status page/email) ______
- Post-incident review checklist:
  - Root cause category: model / data / integration / permissions / policy
  - What guardrail failed?
  - What test would have caught it?
  - What policy change prevents recurrence?

7) Monetization mapping
- What value metric matches the customer’s ROI?
  - Per resolved ticket / per invoice / per lead / per incident
- Pricing sketch:
  - Base fee (governance + integrations): $____/month
  - Usage fee (per resolved outcome): $____
  - Autonomy premium (Tier C): +$____ or +____%
- Monthly value report fields:
  - Outcomes resolved: ____
  - Hours saved: ____
  - Error rate / rollback count: ____
  - Estimated $ impact: ____

If you can fill this out in one sitting and defend the targets to a skeptical CFO or security lead, you’re ready to ship an agentic workflow that lasts.