Roles and Responsibilities in the Operational Phase

Overview
In the PMI Cognitive Project Management for AI (CPMAI) methodology, the final phase, Operationalize the AI Solution, moves a validated model out of the lab and into real business use. This is where many AI initiatives fail. The cause is rarely the algorithm. More often, nobody clearly owns the model once it is live. Understanding Roles and Responsibilities in the Operational Phase means knowing who deploys, monitors, maintains, retrains, governs, and eventually retires an AI solution, and how those people work together.

Why It Is Important
AI systems are not like traditional software. A conventional application keeps behaving the same way until someone changes its code. An AI model can degrade on its own because the world changes around it. Examples include data drift (input data changes), concept drift (the relationship between inputs and outcomes changes), and new regulations. Without clearly assigned roles, several problems appear:
- Model performance silently degrades and no one notices.
- Retraining happens late, too often, or never.
- Bias, ethics, and compliance issues go unaddressed.
- Business users lose trust because there is no feedback channel.
- Accountability gaps emerge when something goes wrong.
CPMAI stresses that AI projects are iterative and lifecycle-based. The operational phase is not the end. It is the start of continuous value delivery. Clear roles keep the AI solution reliable, trustworthy, compliant, and aligned with the business objectives defined in Phase I (Business Understanding).

What It Is
Roles and responsibilities in the operational phase define which people and teams own each activity needed to run an AI system in production. Typical activities include:
- Deployment and integration into business processes and IT systems.
- Monitoring model performance, data quality, and system health.
- Detecting and responding to drift.
- Model versioning, retraining, and redeployment (the MLOps pipeline).
- Governance, ethics, explainability, and regulatory compliance.
- User support, training, and change management.
- Measuring business KPIs and ROI.
- Deciding when to iterate (return to earlier CPMAI phases) or retire the model.

Key Roles and Their Responsibilities
1. AI Project Manager / CPMAI Practitioner: Coordinates the transition from development to operations and keeps the solution tied to business goals. Facilitates communication across teams and defines handoffs. Plans for ongoing iteration and makes sure operational ownership is assigned before go-live. In the exam, this role is usually the coordinator and facilitator, not the person who personally retrains models.

2. Business Owner / Product Owner / Sponsor: Accountable for business value. Defines success criteria and KPIs and approves changes. Funds ongoing operations and decides whether the solution keeps delivering ROI. Owns the go/no-go and retirement decisions from a business perspective.

3. Data Scientists / ML Engineers: Analyze model performance and diagnose drift. Design retraining strategies, tune models, and validate new versions before release. Also responsible for explainability analysis.

4. MLOps / DevOps Engineers: Build and maintain the deployment pipelines (CI/CD/CT, where CT is continuous training) and infrastructure. Handle model versioning, rollback, scalability, latency, and automated monitoring and alerting.

5. Data Engineers: Keep production data pipelines reliable and maintain data quality, freshness, and availability. Make sure the data used for retraining matches production reality.

6. IT Operations / Infrastructure / Security: Responsible for system uptime, security, access control, disaster recovery, and integration with enterprise systems.

7. Data Governance, Compliance, Legal, and Ethics Officers / AI Governance Board: Oversee responsible AI practices, including bias monitoring, privacy, audit trails, regulatory adherence, and documentation. Approve high-risk changes.

8. Subject Matter Experts (SMEs) and End Users: Provide domain feedback and validate outputs in context. Flag anomalies and, in human-in-the-loop systems, supply corrections or labels that feed retraining.

9. Change Management and Training Leads: Drive adoption, train users, and manage organizational resistance. Communicate how the AI is used and what its limitations are.

How It Works
Step 1: Plan operational ownership early. CPMAI encourages operational planning to begin before deployment, ideally from Business Understanding onward. Roles are documented with tools such as a RACI matrix (Responsible, Accountable, Consulted, Informed).
Step 2: Formal handoff. The development team transfers models, documentation, model cards, monitoring thresholds, runbooks, and retraining procedures to the operations owners.
Step 3: Continuous monitoring. MLOps tooling tracks technical metrics such as accuracy, precision, latency, and drift statistics. The business owner tracks business KPIs. Each alert has a defined owner and escalation path.
Step 4: Feedback loops. Users and SMEs report issues. Data scientists investigate. Governance reviews ethical or compliance concerns.
Step 5: Retraining and iteration. When thresholds are breached, the defined owners trigger retraining or loop back to earlier CPMAI phases (Data Understanding, Data Preparation, Model Development, Model Evaluation). The new model is validated and redeployed with version control.
Step 6: Governance and review cadence. Regular reviews confirm the solution remains valuable, fair, and compliant.
Step 7: Retirement. When the model no longer delivers value, the business owner decides to decommission it and IT and MLOps execute the shutdown.

Common Scenarios
- Accuracy drops after a market change. The data scientist diagnoses drift, MLOps retrains and redeploys, and the PM coordinates and communicates.
- A bias complaint arises. The governance/ethics function leads the review, with data scientists supporting the analysis.
- Users ignore AI recommendations. This is a change management and trust issue. The PM and change lead address training, explainability, and communication.
- Production data pipeline breaks. Data engineering owns the fix.

Exam Tips: Answering Questions on Roles and Responsibilities in the Operational Phase
1. Think lifecycle, not project end. Correct answers treat deployment as the start of continuous operation. Be wary of answers implying the project is done once the model is deployed.
2. Prefer answers that assign clear ownership. If a question describes a problem caused by ambiguity, the best answer is usually to define roles, create a RACI, or establish an accountable owner.
3. The PM coordinates; specialists execute. When asked what the project manager should do, choose facilitation, communication, escalation, or alignment with business goals. Do not choose hands-on technical tasks such as personally retraining the model.
4. Business owner equals accountability for value. Questions about ROI, continuing or retiring the model, or KPI alignment point to the business owner or sponsor.
5. Drift questions point to monitoring plus retraining. The best answer usually involves monitoring metrics, diagnosing the cause, and iterating back through CPMAI phases. Avoid answers that ignore the issue or simply build a new model from scratch without analysis.
6. Ethics and compliance go to governance. For bias, privacy, or regulatory concerns, look for answers involving the governance board, compliance, or ethics review, along with transparency and documentation.
7. Plan early. CPMAI favors answers that define operational roles and monitoring plans before go-live, not after a failure.
8. Include humans in the loop. Answers that use SME and end-user feedback to improve the model are typically stronger than purely automated options for high-risk use cases.
9. Watch for adoption issues. Low usage is often an organizational problem. Pick change management, training, and trust-building over technical tweaks.
10. Eliminate extreme answers. Options such as 'shut the model down immediately' or 'let the vendor handle everything' are rarely correct unless the scenario signals critical harm.
11. Remember CPMAI iteration. The methodology is iterative and agile-like. Returning to earlier phases is normal and expected, not a failure.

Quick Memory Aid: 'DMRGA'
- Deploy (MLOps/IT)
- Monitor (MLOps, data scientists, business owner)
- Retrain (data scientists, data engineers)
- Govern (compliance, ethics board)
- Adopt (change management, users, PM)

Summary
Operationalizing AI succeeds when every lifecycle activity has a named, accountable owner. Monitoring, retraining, governance, and adoption are continuous responsibilities shared among business, technical, and governance roles, with the project manager orchestrating alignment. On the exam, choose answers that show clear accountability, continuous monitoring, early operational planning, responsible AI governance, and iterative improvement tied to business value.