Product
Outcomes, scope, quality and feedback.
Library / 07 Implementation, Adoption & Change
Reviewed Oct 2026AI Pilot to Production
A successful demonstration is not yet a supportable service.
01 / flow
Lifecycle
DiscoveryScoped pilotReadiness gatesLimited rolloutOperations + review
02 / table
Release gates
| Gate | What must hold |
|---|---|
| Quality | Representative evals, high-severity bars, risk cases |
| Authority | Approvals, tool scopes, audit, pause and rollback |
| Reliability | Timeouts, duplicates, retries and recovery tested |
| People | Named owners, escalation, support and training |
| Economics | Cost per successful outcome, including human review |
03 / cards
Who owns the rollout?
Engineering / ops
Deployments, instrumentation, incidents and recovery.
Risk / security / data
Access, policy, retention and audit.
Business
Adoption, enablement and staffing.
04 / note
Worked example
A pilot used by 20 employees is not automatically ready for 2,000. Retest permissions, languages, peak load, support, costs and failure ownership.
05 / list
Ask before launch
- What blocks release, and who can disable it?
- Who maintains sources and evals after the pilot?
- How do failures become regression tests?
GOING DEEPER
Scaling is another product problem
20 → 2,000 worked example: A pilot assistant succeeds with 20 internal users. Before deployment to 2,000, verify source permissions under varied roles, support-request volume, peak response time and cost, regional/language differences, capacity, training and the accountable owner when outputs are wrong. A pilot proves feasibility only under its tested conditions.
Gate and owner
| Gate | Named owner / responsibility |
|---|---|
| Quality and outcome | Product: expected behavior, success bars and adoption |
| Security and authority | Risk, data and security: access, auditability, human approval |
| Operational support | Engineering/ops: tracing, incident response, fallbacks, rollback |
| Business readiness | Business lead: training, staffing, workflows and escalations |
| Economics and retirement | Product + business: cost per outcome, ongoing support and a decommissioning route |
Exit rule: A launched feature still needs maintenance, model and vendor-change evaluation, ongoing permissions review, incident readiness, and eventual retirement or replacement.