I've spent the last decade watching machine learning go from academic curiosity to business necessity. But nothing prepared me for the organisational gymnastics I've witnessed over the past 18 months. The way UK unicorns structure their ML teams has evolved dramatically - and frankly, most recruiters I talk to haven't caught up.
The ML Org Chart: Not What You Think
The days of the lone data science team huddled in a corner are over. UK unicorns aren't building isolated ML departments anymore - they're embedding machine learning capability throughout the business.
In fact, the most successful setup I've observed looks nothing like the pyramid structures recruiters typically search for. It's more like a hub-and-spoke model, with a small core ML infrastructure team (usually 5-8 people) serving as the central nervous system, while domain-specific ML engineers are embedded across product teams.
Monzo's approach particularly stands out. Publicly, their engineering blog describes a deliberately small central ML platform team serving a distributed network of ML practitioners embedded across product verticals — a hub-and-spoke model that has become a template for other UK fintechs. Each vertical has autonomy but shares common tooling.
This hub-and-spoke structure solves the biggest headache these companies face: bridging the gap between ML research and actual business value. Too many well-built models have stalled in notebooks because no one could deploy them.
Beyond Tick Boxes: Diversity Recruitment Strategies That Actually Transform UK Workplaces
Master the Virtual Hot Seat: 7 Video Interview Techniques Recruiters Don't Tell You
How to Master 'Tell Me About Yourself' Interview Question: UK Expert Insights
Team Size - Less Is Actually More
Contrary to what most people expect, the core ML teams at UK unicorns are surprisingly compact. Companies valued at £1B+ typically maintain central ML teams of 8-12 people.
Makes sense when you think about it. These companies aren't Google or Meta - they don't need hundreds of researchers pushing the boundaries of what's possible. They need small, elite teams who can apply existing techniques to specific business problems.
The exception? Organisations whose entire product is AI-driven. Take Tractable - their computer vision system for assessing vehicle damage is their product, so naturally their ML team runs larger, closer to 25-30 specialists focused on both research and productionisation.
But for most UK unicorns, the ML team sweet spot hovers around 10 core people, with perhaps another 15-20 ML-capable engineers distributed across product teams.
Reporting Lines: Where ML Actually Sits
Three years ago, most ML teams reported through Data. Three years back, most ML teams reported through Data. Today, they're increasingly reporting through Engineering or Product.
Why the shift? Simple. Machine learning isn't just about building models anymore - it's about building products that happen to use ML. Closer alignment with engineering ensures models actually make it to production. Closer alignment with product ensures they solve real business problems.
Gousto, for instance, restructured last autumn so their ML team reports directly to the CTO rather than the Head of Data. Their reasoning was pragmatic - ML features were getting stuck in the handoff between data science and engineering. The new structure eliminates that bottleneck.
But, and this is crucial, the most sophisticated orgs maintain a dotted line to the Chief Data Officer for governance and data quality purposes. ML without good data is just fancy guesswork.
Research vs Deployment: The Balance Has Tipped
The reality is that most ML techniques UK unicorns need already exist. The challenge isn't inventing new algorithms - it's applying and operationalising the ones we have.
This reality is reflected in team structure. Typically, I'm seeing a 20/80 split between research and application/deployment. Even that's generous - some companies run closer to 10/90.
The days of maintaining a pure research team are largely gone for all but the biggest players. Babylon Health, which filed for administration in August 2023 and ceased UK operations, serves as a cautionary example of the risks of building large research teams disconnected from deployment realities and commercial sustainability.
This doesn't mean research isn't happening - it just means it's focused and practical rather than open-ended. Companies like Benevolent AI still maintain a stronger research emphasis given their domain, but they're the exception rather than the rule.
What They Hire For First
When a UK unicorn starts building their ML function from scratch, the first three hires follow a consistent pattern:
- ML Engineer with strong productionisation experience
- ML Platform Engineer focused on infrastructure
- Applied ML Scientist with domain expertise
Noticed what's missing? The pure researcher. The academic superstar with the PhD but no deployment experience. That hire comes later - if at all.
The other striking trend is the preference for T-shaped profiles over specialists. Companies want people who can contribute across the ML lifecycle - from data preparation to model development to deployment and monitoring.
I placed a Lead ML Engineer at a fintech unicorn last month who told me: "Three years ago they would have asked me about the latest reinforcement learning techniques. This time the interview was all about how I'd managed model governance, testing, and deployment pipelines."
Hiring for Culture Fit
Beyond technical skills, there's an increasing emphasis on culture fit - specifically finding ML practitioners who can communicate effectively with non-technical stakeholders.
I've watched numerous brilliant technicians fail because they couldn't explain their work to product managers or executives. The unicorns that succeed hire for technical excellence AND communication skills.
The other quality they prize? Pragmatism. UK unicorns consistently favour practitioners who can balance the perfect against the good enough. As one CTO told me, "I'd rather have a simple model in production than a brilliant one in a Jupyter notebook."
Size Matters: Different Structures at Different Stages
It's worth noting the ML team structure evolves as companies grow. Pre-Series B, you'll typically see a generalist ML engineer who does everything. By Series C, a small dedicated team forms. Full unicorn status usually brings the hub-and-spoke model I described earlier.
This progression isn't just about headcount. It reflects the maturing AI strategy within the organisation. Early-stage companies experiment. Growth-stage companies establish foundations. Mature unicorns scale and embed ML throughout the business.
What This Means For Recruiters
If you're recruiting for ML roles at UK unicorns (or aspiring unicorns), the implications are clear:
- Look beyond the central ML team - many of the most impactful roles sit within product verticals
- Prioritise deployment experience over pure research credentials
- Value candidates who understand the full ML lifecycle, not just model building
- Don't underestimate the importance of communication skills
Look for specialists who can go broad when needed. The ML Engineer who can also tackle data engineering challenges. The ML Research Scientist who understands what it takes to deploy a model. The unicorns themselves are looking for these hybrid profiles - your shortlists should reflect that.
The ML landscape continues to evolve rapidly. What works today might not work tomorrow. But right now, UK unicorns are finding success with small, elite core teams serving a distributed network of domain-specific ML practitioners. Structure your hiring strategies accordingly.
The jobs section at The OHub shows this trend in action - more hybrid ML roles appearing every week, especially in fintech and healthtech. Worth a look if you're tracking the evolution of ML team structures in real time.
