🎯 Delivery — Objectives¶

🎯 Objective¶

The primary objective of the Delivery phase is to transition the AI system from development to production in a controlled, safe manner — ensuring technical integration, human oversight, user adoption, and regulatory compliance. This is not a "flip the switch" moment. Go-live is a planned sequence of technical deployment, user training, monitoring activation, and compliance verification.

We embed a culture of Human-in-the-Loop oversight and Red Button accountability: employees are rewarded for reporting errors, and psychological safety is central. The AI assists, but the human retains final responsibility. This culture must be established during Delivery, not retrofitted after incidents occur.

Key result: An operational AI system that is technically integrated into the production environment, governed by human oversight procedures, accepted by users, and documented in a complete compliance dossier ready for regulatory audit.

✅ Entry Criteria (Definition of Ready)¶

Before this phase starts, the following conditions must be met:

• The Development phase is completed with Gate 3 (Production-Ready) approved.
• All automated tests have passed on the release candidate configuration.
• The infrastructure for Go-live is ready: production environment, monitoring dashboards, alerting pipelines, and access management are configured.
• The implementation team is on standby and has a documented Go-live plan with rollback procedures.
• The intended Collaboration Mode is confirmed — human oversight procedures are designed for the specific mode.

⚙️ Core Activities¶

1. Technical Integration¶

We connect the AI system to the existing software architecture, security infrastructure, and operational monitoring.

• System Connections: Integrate the AI solution into the current IT architecture. This includes API connections, database integrations, authentication/authorisation, and network configuration.
• Access Management: Set up who may use which functions and data. Implement role-based access control (RBAC) aligned with the organisation's identity management system.
• Stability Test: Confirm that the integration does not cause disruptions to other processes. Run load tests, stress tests, and failover tests before Go-live.

2. Human Oversight¶

We implement human supervision procedures as required for the chosen Collaboration Mode. Human oversight is not a generic concept — it is a set of concrete procedures tailored to the mode.

• Oversight Protocols: Record how and when a human must intervene. For Mode 2 (Advisory): the human reviews every AI suggestion before acting. For Mode 4 (Delegated): the human reviews escalations and performs periodic sampling.
• Escalation Paths: Define who is notified when the system operates outside its boundaries. Include contact information, response time expectations, and escalation levels.
• Intervention Levels: Establish clear agreements on the degree of autonomy. What can the AI do independently? What requires human approval? What is prohibited?

3. Adoption & Training¶

We train users not only in the technical operation, but in the new way of working with AI. Adoption is the difference between a system that works and a system that delivers value.

• Workflow Training: Show users how daily work changes with the AI assistant. Demonstrate the before/after comparison so users understand the value.
• Quality Awareness: Teach users how to critically evaluate the AI's output. Users must understand that AI output is probabilistic — it can be wrong, and they are responsible for verifying it.
• Feedback Loop: Set up a channel for user experiences and improvement points. User feedback is the primary input for the Monitoring phase's continuous improvement cycle.

4. Compliance Dossier¶

We complete all documentation required for laws and regulations. The compliance dossier is the evidence pack that demonstrates the system meets the EU AI Act and other applicable regulations.

• Legal Dossier: Collect all reports for the EU AI Act, GDPR, and sector-specific regulations. Include the Risk Pre-Scan, Validation Reports, Technical Model Card, and Guardian approvals.
• Accountability Evidence: Demonstrate that the Hard Boundaries have been maintained throughout testing. Show the audit trail from Goal Definition through Steering Instructions to Validation Results.
• Handover Logs: Complete overview of the system history — every change, every validation, every decision. This is the traceability record that auditors will request.

5. Red Button Culture¶

We establish a culture where employees are rewarded for reporting errors and psychological safety is central. This is not a technical activity — it is an organisational one, but it is critical for safe AI operation.

• Error Reporting Procedure: Set up a simple, accessible channel for reporting AI errors. Make it clear that reporting an error is valued, not punished.
• Response Commitment: Commit to reviewing every reported error within a defined timeframe. Close the loop with the reporter — let them know what was done.
• Leadership Endorsement: Have leadership communicate that the Red Button culture is a priority. When leaders model error reporting, the organisation follows.

👥 RACI¶

✅ Exit Criteria (Gate 4 — Go-live)¶

The Delivery phase closes when all of the following are satisfied:

• Technical integration is complete and stability tests have passed.
• Human Oversight protocols are documented and communicated to all relevant roles.
• User training is completed and feedback channel is active.
• Compliance Dossier is complete and signed off by the Guardian and Business Sponsor.
• Monitoring dashboards and alerts are active in the production environment.
• Rollback procedure is tested and documented.
• Gate 4 review is conducted with the Business Sponsor.
• Go-live decision is documented and communicated.

Collaboration Mode: [Mode X — Name]. Handover protocol specifies human oversight per mode. Required validation for this mode: → See Evidence Standards.

📦 Deliverables¶

The following artefacts are produced during this phase:

1. Go-live Plan — deployment sequence, rollback procedure, and communication plan.
2. Human Oversight Protocol — intervention procedures, escalation paths, and autonomy levels.
3. Training Materials — workflow guides, quality awareness content, and feedback channel documentation.
4. Compliance Dossier — complete evidence pack for regulatory audit.
5. Operational Handover Checklist — confirmation that all production requirements are met.

Next step: Complete the handover checklist and activate the monitoring dashboard. → Use the Operational Handover Checklist as your starting point. → See also: Activities | Gate 3 Checklist | Traceability

Version: 1.1 Date: 07 May 2026 Status: Final