Lasting success with AI comes not from individual projects but from a well-built team. Organisations building their first AI team tend to face the same questions: which roles do we need? Whom should we hire first? Where should the team sit?
Core roles
| Role | What they do |
|---|---|
| AI product owner | Defines and prioritises business problems, bridges to business units |
| Data scientist | Analyses data, builds and validates models |
| ML engineer | Turns models into production-quality code and serves them |
| AI / LLM engineer | Builds LLM-based applications (RAG, agents) |
| Data engineer | Builds data pipelines, makes data accessible and reliable |
| MLOps / platform engineer | Builds and runs training, deployment and monitoring infrastructure |
| Domain expert | Brings business knowledge: credit risk, customer behaviour, operations |
| Governance and risk lead | Manages regulation, ethics, model risk and data privacy processes |
In a small team some of these roles may be combined in one person. What matters is that someone owns each responsibility.
In what order should you hire?
A common mistake is to hire a few data scientists first and expect results. If data and infrastructure aren't ready, data scientists spend most of their time looking for data and failing to get models into production.
A usually more effective sequence:
- Team lead / product owner: prioritisation and relationships with business units
- Data engineer and data scientist: to deliver the first use case end to end
- ML / MLOps engineer: once several models start going into production
- LLM engineer: when GenAI use cases enter the roadmap
- Governance lead: as AI use spreads and regulatory requirements grow
Operating models
Centralised
The whole AI team sits in one unit (e.g. an AI centre of excellence) and serves the business units.
- Pros: standards, shared infrastructure, knowledge sharing
- Cons: distance from the business, risk of becoming a bottleneck
Federated
Each business unit has its own data and AI team.
- Pros: closeness to the business, speed
- Cons: duplicated work, inconsistent standards, no shared infrastructure
Hybrid (recommended for most organisations)
Platform, standards, governance and deep expertise sit at the centre; business units have domain-focused teams, or specialists who report to the centre but work with the units.
- Pros: both standards and closeness to the business
- Cons: requires clearly defined roles and responsibilities
Process and method
The right process matters as much as the right people:
- AI projects are more uncertain than classic software projects; plan for experiment–evaluate cycles.
- Every project should start with a business metric, a business owner and a plan for production.
- Code, data and models should be versioned, and models monitored in production.
Build or partner?
Not every organisation needs a full in-house AI team. For many, the fastest and most efficient route is a core internal team (a product owner and a few technical specialists) working with external expertise. The internal team grows over time and knowledge stays in the organisation.
Conclusion
A good AI team is built with the right roles, the right hiring sequence and the right operating model.
Through our Team & Process Design service we design your operating model, role definitions and hiring plan with you. To develop your existing teams, see our Data Science with Python, Applied Machine Learning and MLOps programmes.
Sources