Every organisation that takes AI seriously eventually asks the same question: who is going to do this work? The technology is the easy part. The hard part is building a team with the right mix of technical skills, business understanding, and the ability to ship things that actually work in production.
We've helped dozens of organisations build their AI capabilities — from two-person startups to enterprise teams of thirty. The patterns that work are remarkably consistent, regardless of size.
The Key Roles You Need
Not every AI team needs every role from day one. But understanding the full landscape helps you plan. Here are the roles that matter most:
AI/ML Engineer. This is the core technical role. They build, fine-tune, and deploy models. They understand the difference between a working prototype and a production system. In 2026, the best AI engineers are fluent in both traditional ML (scikit-learn, PyTorch) and LLM orchestration (LangChain, LlamaIndex, custom agent frameworks). Finding people who can do both is hard and expensive, but it's worth the investment.
Data Engineer. AI is only as good as its data. Data engineers build the pipelines that get data from source systems into the formats AI models need. They handle data quality, transformation, and infrastructure. Without a solid data engineer, your AI team will spend 80% of their time fighting data issues instead of building models.
Product Manager (AI). Someone needs to decide what to build and why. An AI product manager bridges the gap between business stakeholders who have problems and technical teams who have solutions. They define use cases, prioritise the backlog, set success metrics, and make sure the team is solving problems that actually matter to the business.
AI Ethics and Governance Lead. As AI becomes more embedded in business processes, someone needs to own the governance side: bias testing, compliance with regulations, acceptable use policies, and risk assessments. In smaller teams, this might be a shared responsibility. In larger organisations, it's a dedicated role. Either way, it cannot be an afterthought.
Domain Experts. Often overlooked, but critical. The people who understand the business processes you're automating are as important as the people who build the models. A claims processing AI built without input from actual claims handlers will miss edge cases that tank accuracy in production.
Build, Hire, or Outsource?
This is the decision most organisations get wrong. The instinct is usually to hire a full team immediately, but that's rarely the right move. Here's how we recommend thinking about it:
Outsource first. If you're early in your AI journey, start with an external partner. A good AI consultancy (like us) can deliver your first two or three use cases, establish best practices, and help you understand what skills you actually need in-house. This is faster and lower-risk than hiring a team to figure it out from scratch.
Hire strategically. Once you have proven use cases and understand your needs, start hiring. But hire for the roles that create the most leverage. Usually, that's a strong AI/ML engineer and a data engineer. These two roles can handle a surprising amount of work if your infrastructure is solid.
Build a Centre of Excellence. As your AI programme matures, consider establishing an AI Centre of Excellence — a central team that sets standards, provides shared tools and infrastructure, and supports business units in adopting AI. This prevents the fragmentation that happens when every department builds its own AI capability independently.
Hybrid is usually best. Most mature organisations end up with a mix: an in-house core team that owns strategy, governance, and critical workloads, supplemented by external specialists for spikes in demand, niche expertise, or new capability areas.
Team Structure Patterns That Work
We see three common patterns, each suited to different stages of maturity:
The embedded model. AI engineers sit within business units (e.g., one in marketing, one in operations). This works well for organisations with a few specific use cases, because the engineers develop deep domain knowledge. The downside is siloed tooling and inconsistent practices.
The centralised model. A single AI team serves the entire organisation. They take requests from business units, prioritise them, and deliver. This creates consistency and shared infrastructure, but can become a bottleneck if demand outstrips capacity.
The hub-and-spoke model. A central AI team sets standards, provides shared infrastructure, and handles governance. Business units have their own embedded AI resources for day-to-day work, but they follow the central team's guidelines and use shared tooling. This is the model we recommend for most mid-to-large organisations. It balances speed with consistency.
Upskilling Your Existing Staff
You don't need to hire everyone from outside. Some of your best AI team members are already on your payroll — they just need the right training.
Software engineers can transition into AI/ML roles with targeted training in model development and deployment. The engineering fundamentals transfer directly.
Data analysts already understand data quality, transformation, and business context. With additional training in ML concepts and tooling, they can move into data engineering or junior ML roles.
Business analysts and product managers can become AI product managers with training in AI capabilities, limitations, and evaluation methods. Their existing stakeholder management skills are invaluable.
We offer tailored AI training programmes that are designed for exactly this: helping existing staff build AI skills without starting from zero. The key is making training practical and role-specific, not generic. A workshop on prompt engineering is more useful for a marketing team than a lecture on neural network architectures.
Common Pitfalls to Avoid
Hiring too senior too early. Bringing in a Chief AI Officer before you've delivered a single AI project is putting the cart before the horse. Start with builders, not strategists.
Ignoring the data side. Organisations hire ML engineers and then wonder why nothing gets built. The answer is almost always that the data infrastructure is not ready. Hire data engineers first, or at least in parallel.
No clear mandate. An AI team without a clear remit, budget, and executive sponsor will struggle to deliver. Make sure the team knows what success looks like and has the authority to make decisions.
Treating AI as IT. AI is not a technology project — it's a business transformation initiative. Parking the AI team inside IT often limits their access to business stakeholders and reduces their impact. The best AI teams report to someone with cross-functional authority.
At Grove AI, we help organisations build AI capabilities that last — whether that means delivering your first use cases, establishing a Centre of Excellence, or training your existing team. If you're figuring out how to build your AI function, book a free strategy call and we'll help you plan your approach.