Leading AI Transformation: What the Role Actually Looks Like

“Head of AI Transformation” is one of those job titles that didn’t exist five years ago, and still sounds like something you’d find on a satirical LinkedIn bingo card. I know this because when I tell people what I do, I can see them trying to work out whether I’m serious. Fair enough. I had the same reaction the first time I saw the title on a job spec.

The Title Problem

“AI Transformation” sounds like something someone invented at a management consultancy after too many espressos. I know this. I’ve made the jokes myself. But the role exists because the problem exists: most organisations know they should be using AI more effectively and have no coherent strategy for doing so.

The difference between “we use AI” and “we are AI-native” is the difference between having a gym membership and being fit. One is a purchase; the other is a practice. My job is to build the practice.

What the Day-to-Day Actually Looks Like

A typical week includes:

Monday: Review AI feature pipeline with product leads. 6 features in various stages. About 90 minutes of actual meetings, plus 30 minutes of “quick questions” that are never quick.
Tuesday-Wednesday: Working sessions with teams. This is where the actual work happens. Pair-reviewing AI implementations, unblocking teams stuck on prompt engineering or model selection, reviewing AI Impact Assessments.
Thursday: Strategic work. Roadmap updates, vendor evaluations, writing the materials that make everything else work. This is also when I take the meetings I can’t avoid (vendor demos, cross-functional planning, the occasional board preparation).
Friday: Community of Practice session (every other Friday). Writing. Catching up on what I’ve missed.

If I’m honest, about 40% of my time is strategic, 40% is hands-on with teams, and 20% is administrative. I’d prefer 30/50/20, and I’m working on it.

Enablement vs Mandates

The single most important lesson I’ve learned in this role: enablement beats mandates every time.

You can mandate that every team must use AI in their workflow. You’ll get compliance. People will technically use AI tools. They’ll generate output that technically came from an AI. And none of it will be any good, because people who are forced to use tools they don’t understand will use them badly.

Or you can enable. Show people what’s possible. Give them safe spaces to experiment. Celebrate early wins. Make the tooling easy and the guardrails clear. Let adoption happen because people genuinely see the value.

Our Community of Practice is at 52 members now (up from 35 three months ago). Nobody was mandated to join. Nobody was mandated to use AI tools. But when you see your colleague ship a feature in half the time, you get curious. I’ll take curiosity over compliance.

(A tangent: this is exactly what I got wrong at my first startup. At SolarPrint, Colm and I tried to mandate adoption of new manufacturing processes. The engineering team resisted. When we switched to letting one team pilot the new process and share their results, the other teams adopted voluntarily within a month. I was 26 and learned the hard way that engineers don’t respond well to being told what to do. Neither, it turns out, does anyone else.)

The Strategic Roadmap

I own the AI strategic roadmap, which is a fancy way of saying I maintain a document that answers three questions:

Where are we now? (current AI maturity, measured quarterly)
Where do we want to be in 12 months? (specific, measurable targets)
What needs to happen to get there? (projects, resources, dependencies)

The current roadmap has 4 workstreams: internal productivity (multi-agent workflows, code assistance), product AI (patient-facing and clinician-facing features), data and infrastructure (the platform work that makes everything else possible), and capability building (training, CoP, standards).

Each workstream has an owner, a quarterly target, and a budget. The total incremental budget for AI transformation this year is, let’s say, modest by Silicon Valley standards but meaningful for a healthcare company of our size. Most of the cost is people’s time rather than technology spend. The tooling is surprisingly affordable; Claude Pro costs about €20/month per seat. The expensive part is the organisational change.

Driving Adoption

The adoption curve in any organisation follows a predictable pattern. You have your early adopters (about 15% of the company, in my experience). They’ll try anything new and are usually too enthusiastic. Then you have your early majority (35%), who adopt when they see proof it works. Then your late majority (35%), who adopt when it becomes the default. And your holdouts (15%), who adopt when they have no choice.

I’m currently working on the early majority. The early adopters are already converted. The early majority needs evidence, case studies, and low-risk entry points. That’s why I track everything: time saved, quality improvements, incident rates. Not because I love spreadsheets (I absolutely do not), but because the early majority makes decisions based on data, not enthusiasm.

The most effective thing I’ve done for adoption, though, isn’t data. It’s pairing. I sit with teams and work through real problems using AI tools. Not hypothetical demos. Their actual work, their actual codebase, their actual constraints. Two hours of pairing converts more people than any presentation I’ve ever given.

The Honest Bit

I don’t have this figured out. I’m 6 months into leading AI transformation at Oneview, having evolved from the Technical PM role I started in March, in an industry that moves slower than tech but faster than it used to. I make mistakes regularly. Last month I pushed a workflow on a team that wasn’t ready for it and had to walk it back. The month before, I underestimated a compliance requirement and cost a project two weeks.

But I’ve been in AI since before it was trendy. I did my PhD in computational modelling. I built a smart buildings startup. I was working with neural networks when the best models could “sometimes recognise cats.” So when someone asks me if this is a fad, I can say with some confidence: it’s not, and the organisations that build the practice now will have a meaningful advantage in three years.