Roles
Why roles are needed
An AI initiative is almost never implemented by one team. Even when it seems like "just connect an LLM" or "build RAG," the initiative touches business process, data, AI products, IT integrations, security, operations, users, and impact measurement.
That is why the portfolio needs clear accountability upfront. Otherwise business waits for results, IT waits for requirements, the AI team waits for access, security joins too late, and no one can confirm impact.
Main principle: every initiative must have an outcome owner. The AI function helps move the initiative through the system, but should not own every business effect.
Core roles
1. Business sponsor
Owns the business problem behind the AI initiative: what problem is solved, why it matters, expected impact, users, process adoption, and who confirms usefulness.
They do not need to know whether the solution is LLM, RAG, ML, agent, or automation. Their job is to describe the need and confirm value.
Main responsibility: business meaning, impact, and adoption into the real process.
2. AI initiative lead
Drives the initiative through the business funnel from idea to closure. Clarifies the problem, frames value hypothesis, identifies candidate AI products, organizes feasibility assessment, gathers participants, manages statuses, prepares stage gate materials, records decisions and risks, and prevents the initiative from getting stuck.
Main responsibility: move the initiative through the business funnel and bring it to realized impact.
3. AI product owner
Owns the AI product through which the initiative may be implemented: LLM, RAG platform, ML platform, code agent, applied AI service, etc.
They determine whether the product fits the task, what constraints exist, what data/configuration is needed, whether a configuration/prompt/RAG base/agent is enough, what product improvements are required, and how the initiative affects the product roadmap.
Main responsibility: connect the business need to a suitable AI product or show that current products are insufficient.
4. Impact owner
Business-side person who confirms that the initiative delivered a result. This can be the business sponsor, but not always: it can be a process owner, finance controller, or metric owner.
They own baseline, target metric, confirmation method, actual usage, and confirmation of savings, revenue, effort reduction, process acceleration, or quality improvement.
Main responsibility: confirm measurable or qualitatively recorded impact.
5. Users / working group
Real users who test the solution in work. They confirm usability, real task fit, insufficient quality, uncovered scenarios, usage risks, and changes needed before scaling.
Main responsibility: provide practical feedback and confirm applicability in real work.
6. Adjacent functions
Depending on initiative type and risk level, adjacent functions include IT, architecture, information security, data owners, DWH/data teams, legal, compliance, operations, and support.
Their job is not to block, but to check readiness for enterprise use: access, data, integrations, architecture, security, operations, regulatory constraints, and post-launch support.
Main responsibility: enable safe and sustainable adoption.
Additional governance and control roles
| Role | Why it is needed |
|---|---|
| Initiator | Frames the initial problem and starts the initiative card |
| Data owner | Owns sources, quality, and permission to use data |
| Security | Checks access, confidentiality, and processing perimeter |
| Architect | Checks integrations, operations, observability, and rollback |
| Finance | Helps confirm meaningful economic impact |
| Portfolio committee | Decides priorities, resources, exceptions, and disputed transitions |
Not every role is needed from day one, but before delivery the business owner, movement owner, selected product, and data/security constraints must be clear.
Role of the AI function
The AI function should not be the only executor of all AI initiatives. It is a management and methodology center that helps the business move from idea to impact.
It owns initiative intake rules, business funnel, product routing, prioritization, stage gates, status transparency, participant synchronization, artifact control, reusable solution accumulation, and impact recording.
The AI function should not own impact instead of business, approve every detail manually, replace security/architecture/finance, or push ownerless initiatives into delivery.
Role map
| Role | Responsibility |
|---|---|
| Business sponsor | Problem, value, process adoption |
| AI initiative lead | Movement through business funnel |
| AI product owner | Selection and evolution of the right AI product |
| Impact owner | Result and benefit confirmation |
| Users / pilot group | Applicability check in real work |
| Adjacent functions | Data, IT, security, architecture, operations |
| AI function | Portfolio management, rules, stage gates, transparency |
Responsibility matrix
| Action | Main role | Participants |
|---|---|---|
| Create initiative | Initiator | AI assistant, AI function |
| Clarify problem and impact | Business sponsor | Initiator, AI function |
| Check similar initiatives | AI function | AI assistant |
| Select product | AI function | AI product owner, architect |
| Check data | Data owner | Security, delivery team |
| Check security | Security | AI function, data owner |
| Prepare delivery | AI initiative lead | AI product owner, delivery team |
| Check architecture | Architect | AI product owner, security |
| Confirm impact | Impact owner | Finance, AI function |
| Approve exception | Portfolio committee | Sponsor, AI function, risk functions |
Minimum roles by stage
| Stage | Required roles |
|---|---|
| New | initiator, AI function |
| Evaluation | business sponsor, AI function, AI product owner if route exists |
| Delivery | AI initiative lead, AI product owner, data owner, security when needed |
| Awaiting impact | impact owner, AI function, finance for material impact |
| Support | AI product owner, operations owner, business sponsor |
AI assistant role
The AI assistant helps roles work faster: initiator with brief, AI function with completeness and duplicate checks, finance with draft impact model, security with initial questions, architect with integration description, initiative lead with tasks.
But the assistant is not the decision owner and does not replace access rights. Its actions must follow the same rules as user actions.
Anti-patterns
Bad role model:
- initiative has no business owner;
- AI function owns impact instead of business;
- initiative lead is assigned formally;
- security joins at the end;
- data owner is unknown;
- committee decides without criteria;
- no one owns support after launch.
A good role model makes accountability visible before delivery, not after the first failure.
Core idea
The AI initiative portfolio is not a list of ideas and not scattered pilots. It is a managed loop where each initiative has problem owner, movement owner, product owner, delivery owner, impact owner, approval participants, and clear business funnel status.
That turns AI initiatives from chaotic experiments into a managed way to convert business needs into working AI solutions.