You know your organisation needs to get more from its data and AI. Someone on the leadership team has mentioned hiring. The job boards are full of titles that didn't exist three years ago — AI Engineer, ML Engineer, Prompt Engineer, Head of AI — sitting alongside the familiar ones. And now you're trying to work out which of them you actually need.
This decision trips up a lot of leaders — not because it's inherently complicated, but because the people selling you the answer usually have a stake in which way you go.
Here's a plain-English breakdown of the roles, what they actually do, and when each one makes sense.
Data Analyst
A data analyst works with data you already have to answer questions you already know how to ask. They build dashboards, produce reports, run queries, and help the business make sense of historical and current performance.
If you're still running key metrics out of spreadsheets, making decisions based on reports that take days to produce, or struggling to get a consistent picture of your business across departments — a good data analyst will transform your operational visibility.
What they typically won't do: build predictive models, architect data infrastructure, or lead an AI initiative.
Right for you if: You need better visibility of what's happening in your business right now, and you have data that's reasonably clean and accessible.
Data Scientist
A data scientist builds models — predictive, classification, recommendation, generative. They work at the intersection of statistics, programming, and domain knowledge to extract intelligence from data that isn't accessible through standard reporting.
This is genuinely powerful work. It's also expensive, requires good data infrastructure to sit on top of, and takes time to deliver business value. Most data scientists will tell you (honestly) that 60–80% of their time is spent on data preparation rather than modelling.
The trap: hiring a data scientist before the data foundation is ready. Working with a global publishing organisation, we saw a talented data scientist spend the first six months of their tenure fixing data pipelines and quality issues that predated their hire. The modelling work they were brought in to do didn't start until month seven. That's not a hiring failure — it's a sequencing failure.
Right for you if: You have a well-defined prediction or classification problem, reasonably clean data, and existing infrastructure that a data scientist can build on. Not as a first hire into a team with no data foundation.
AI Engineer
This is one of the newer titles, and one of the most important to understand. An AI engineer builds systems and products that use AI — integrating foundation models, APIs, and services into applications and workflows. They're closer to a software engineer than a data scientist: their job is to take AI capabilities and make them work reliably in production.
Think: the person who builds your customer-facing AI assistant, integrates an LLM into your internal tools, or connects your data to a RAG pipeline. They're not typically training models from scratch — they're building with the models that already exist, grounding them in your data, and making them production-ready.
This role has become increasingly central as foundation models (GPT, Claude, Gemini, Llama) have made it possible to build sophisticated AI applications without starting from scratch. The skill is in the architecture, the integration, the grounding, and the reliability — not the model research.
Right for you if: You want to build AI-powered applications or workflows using existing foundation models, and you need someone who can take it to production rather than just prototype.
ML Engineer
An ML engineer sits between data science and software engineering. Their primary job is getting models built by data scientists into production — and keeping them there. They build the deployment pipelines, monitoring systems, retraining automation, and infrastructure that make ML reliable at scale.
The "last mile to production" problem is one of the most common failure modes in AI programmes: great models that never ship, or ship and degrade silently. An ML engineer is the person who solves that.
Right for you if: You already have data scientists building models but nothing is making it into production reliably, or you need to operationalise ML at scale.
Head of AI / AI Lead
This is a strategic and organisational role, not a technical delivery role. A Head of AI defines how AI capability is built across the organisation — what gets built, in what order, with what governance, and how it connects to business outcomes. They work with leadership to set the AI agenda and with technical teams to ensure it's executed coherently.
Most organisations need this role earlier than they think — not because they're ready to build an AI team, but because without it, AI initiatives scatter across departments without coordination, governance, or a coherent strategy. The first Head of AI often spends more time stopping the wrong things than starting new ones.
Right for you if: AI is becoming a serious strategic priority and you need someone to own the agenda, not just deliver individual projects.
Prompt Engineer
This title has generated more heat than it deserves. In practice, prompt engineering — crafting inputs to get reliable outputs from language models — is a skill that most AI engineers and developers acquire as part of their broader work. It's rarely a standalone role at the scale most organisations operate.
That said, in teams doing high-volume or high-stakes LLM work (legal, compliance, content generation at scale), having someone who focuses on prompt design, evaluation, and systematic testing is genuinely valuable. Just don't hire a prompt engineer as your first AI hire.
Right for you if: You already have an AI engineering capability and need to systematise and optimise your LLM interactions at scale.
Consultant
A consultant brings external expertise for a defined period to solve a specific problem or build a specific capability. The good ones transfer knowledge to your team so you can operate independently afterwards. The not-so-good ones build dependency.
Consultants make most sense when:
- You need to move faster than hiring allows
- The problem is defined enough to scope but complex enough to require specialist experience
- You want to build internal capability rather than create a permanent external dependency
- You need senior expertise across multiple disciplines (data engineering, AI, cloud, change management) that doesn't make sense to hire full-time at your scale
The honest version: a good consultancy engagement should end with your team more capable than when it started, not more reliant on the consultant.
Right for you if: You have a specific outcome you're aiming for, and you want to build capability rather than rent it indefinitely.
The question most leaders still don't ask
Before deciding which type of person or partner you need, it's worth asking a more fundamental question: is the problem I'm trying to solve a data problem, a technology problem, or a business process problem?
Most AI and data initiatives that fail don't fail because of the model or the tooling. They fail because the underlying business process wasn't defined, the data wasn't good enough, or the people who needed to change how they worked didn't. None of the roles above solve those problems directly — and a consultant only solves them if the engagement is scoped to include them.
The right question isn't "who do I hire?" It's "what is the outcome I'm trying to achieve, and what's actually blocking it?"
The shift that changes the calculus
For most of the last two decades, AI and data capability was effectively rationed by headcount. If you couldn't afford a team of engineers, analysts, and scientists, you were limited to the reports your ERP or CRM could produce. Those systems digitised processes but didn't add intelligence — they told you what happened, not what would happen.
That's changed. Agentic AI handles much of the work that used to fill data engineering timesheets. Foundation models have replaced years of model research. Cloud infrastructure and co-funding have reduced the capital barrier. A lean team with the right setup can now produce the kind of intelligence that previously required ten specialists to generate.
This matters for the hire vs consultant question. The answer isn't always "hire someone." It's often "build the right foundation with experienced external support, then your existing team can operate it." With cloud co-funding available for most AI workloads, the ROI of genuine business intelligence — prediction, automation, early warning built on your data — is no longer a large-organisation calculation.
Most leaders who come to us thinking they need to hire a data scientist actually need one of three things first: a clear picture of what data they have and what it could power; a specific, well-scoped problem defined well enough to build against; or a foundation that makes any hire viable and productive.
None of those require a full-time data scientist. All of them move faster with experienced external support.