Review the engagement dossier.

Executive position

Customer-facing copilots is not launch-ready. Treat this as a control-build engagement before any expansion, with executive sign-off and evidence review before release.

Decision rule: 3 stage + 4 use case + 2 timeline + 1 stack + 10 risk = 20 total.

Engagement outputs

1. 01Intake summary with assumptions, constraints, and launch decision criteria.
2. 02AI system inventory covering data sources, model vendors, human review, and owners.
3. 03Prioritized risk register with severity, owner, mitigation, and proof needed.
4. 04Customer journey risk map with launch approval gates.
5. 05Golden-task evaluation pack with acceptance thresholds.
6. 06Data handling review covering retention, redaction, and vendor boundaries.
7. 07Measurement plan for offline evals, review sampling, and production monitors.
8. 08Engagement roadmap with weekly milestones and decision checkpoints.

Control conditions

• Confirm named business, technical, security, and review owners before work starts.
• Define stop/go gates for pilot, launch, and rollback decisions.
• Require golden tasks, reviewer sampling, drift thresholds, and launch-blocking eval gates.
• Map prohibited data, retention limits, redaction ownership, and vendor logging boundaries.
• Define offline eval coverage, production monitors, sample rates, and failure triage SLAs.

Discovery questions

1. Who owns executive sign-off for launch, remediation, or pilot expansion?
2. Which production data can be sampled for evaluation without creating new risk?
3. What customer, compliance, or security evidence must be produced before launch?
4. What internal milestone should this engagement unblock?
5. Which parts of the current stack are mandatory versus replaceable?
6. Which examples define a correct, borderline, and failed response?
7. Which data classes are prohibited from model prompts or logs?
8. What signals will prove the system is safe enough to launch?