A Conversation With Bob De Caux, Chief AI Officer At IFS On Sovereign AI Power Index

Please introduce yourself and tell us about your role at IFS and your experience working in AI.

I’m Chief AI Officer at IFS. My background is in agent-based modelling and how complex systems behave, a lens that still shapes how we approach applied AI.

At IFS, our strategy is centred on industrial AI. I have always been a strong believer that AI has to be embedded deep within a product and when it is surfaced to users, it has to be domain-specific and easy to use.

We don’t start with the technology and ask where AI can help. We start with the industries we serve – the practical, day-to-day problems a technician, a planner or a manufacturer faces and work back to the technology. IFS software is where agents become real: work orders, schedules, parts, the systems that connect a decision to an operational outcome.

You’ve engaged in major policy discussions, including conversations around AI policy in the UK. What first drew you to AI, and what keeps you excited about the technology today?

What drew me to AI was the idea that it could help us answer big, complex questions that we weren’t able to tackle before. While that is still the case, what keeps me on my toes today is that you can’t answer those questions in a vacuum and the practical reality of how you deliver AI-based solutions is still a rapidly developing area. The immediate challenge for many businesses is building practical AI capabilities that can operate reliably.

What is genuinely exciting from an engineering perspective is building the technology backbone to deliver AI capabilities. We are moving towards a thread of intelligence where we can take real-time data feeds from assets, run predictive AI algorithms under the bonnet, and connect intelligent systems across complex business processes. The challenge is making that thread reliable and governed in industrial environments. When you have thousands of processes, systems and sources of information, enabling AI to select the right capability while maintaining strong governance becomes a fascinating structural shift.

Why should UK businesses and startups care about the launch of the Sovereign AI Power Index?

The reality of enterprise implementation is a long way behind the hype train. There’s huge excitement, but most companies are still trying to understand what AI actually means for their operations. The Sovereign AI Power Index is valuable because it shifts the focus away from pure technology ambition onto operational readiness.

For UK businesses and startups, AI leadership is not about rushing to adopt the latest, greatest generic LLM, which would still need to guess its way through a complex task. Leadership is about readiness: the data foundations, governance, infrastructure and organisational maturity to take AI out of the pilot phase and into meaningful, complex industrial workflows, securely.

The index measures countries on areas including compute capacity, capital formation, regulatory readiness, data sovereignty and intelligence capability. Where do you think the UK currently stands weakest, and what would realistically unlock improvement?

The biggest hurdle right now is moving from experimentation to scale. Plenty of organisations are still running experiments, with far fewer having crossed into enterprise-grade deployment. Unlocking that takes architectural discipline. Developing AI features is expensive and requires real infrastructural changes. To scale, you need to automate more of the data science work required to create the models, then get them running and usable in production.

We’ve been building automated pipelines that push the complexity into the background. If the UK wants to improve across the Index’s core dimensions, we need less shallow experimentation and more investment in data readiness, industry-specific knowledge structures, and the infrastructure that translates technical capability into predictable operational value.
 

Is AI sovereignty actually achievable for countries like the UK, or are we really talking about managing long-term dependency on a small number of global technology providers?

I see AI sovereignty as a question of control, resilience and data boundaries rather than complete independence from global technology providers. No country or company is likely to build every part of the AI ecosystem in isolation; the investment and infrastructure required means organisations will continue to work with specialists. The important question is whether you retain control: over how those technologies are applied, where your data sits and how AI is governed within your operations. If you’re an airport, wind farm operator or manufacturer and those things live in someone else’s platform, a supplier’s commercial decision can disrupt critical operations.

That’s why we’re building an intelligence sovereignty architecture at IFS. The reasoning sits on top of the customer’s systems of record and data, inside boundaries they set and can audit, with the model layer kept interchangeable underneath. Customers can draw on the best available models without surrendering control of the things that are genuinely theirs. Sovereignty, in practice, isn’t a wall around your border – it’s keeping ownership of your data, your decisions and your ability to operate, wherever your models come from.

The index is designed to benchmark readiness, not just ambition. What separates countries or organisations that pilot AI from those that scale it successfully in real industrial environments?

It comes down to being outcome-focused rather than chasing the technology for its own sake. If you just throw AI at your business without understanding the problem you’re solving, it just creates more complexity and more systems to manage.

The organisations that do well do not usually make a big-bang transition. They run architectures in parallel – keeping their stable legacy systems running while identifying specific areas where AI can deliver value. Focusing on a clear operational outcome and capturing that in the design makes it much easier to test, govern and scale across the enterprise.

Do you think the biggest constraint on improving a country’s AI score is regulation and infrastructure, or something less discussed like workforce readiness, trust and frontline adoption?

Infrastructure and regulation matter, but workforce readiness, trust and organisational change are just as important. The next phase of AI adoption is not just about building more capable models; it’s about making systems reliable, governed and usable in the real world. In industrial environments, adoption depends on trust and practical usefulness. Field technicians, factory operators and engineers do not need AI for its own sake – they need tools that help them solve problems and make better decisions. That trust comes from governance, auditability and clear boundaries: knowing how a decision is made while maintaining human oversight. It’s why it can help to think of agents as a workforce, not software – they need identity, role-based access, separation of duties and audit trails.

Get the design right and the payoff is tangible: when a technician can use IFS Copilot through natural language to access information or complete a workflow instead of digging through documentation, AI stops being an abstraction and becomes a practical tool that helps them do their job more effectively.

The UK often talks about becoming an AI superpower. What do you think is the biggest obstacle standing in the way of that ambition?

The biggest obstacle might be the definition of “superpower” itself. If it means matching the US or China on frontier models – the largest clusters, the most compute, the most capital – the UK won’t win that race, and the real risk is pouring resources into trying.

The UK’s actual advantage is somewhere else. We have world-class industries, including aerospace, life sciences and financial services, sitting on deep, specialised data and decades of hard-won domain expertise. That’s precisely where applied, industrial AI creates value, and it’s the layer the frontier labs can’t easily reach, because they don’t have the domain knowledge or the trusted data to operate in it. A country that leads in applying AI to hard, regulated, real-world problems is a superpower in the sense that it actually changes economic outcomes.

So the obstacle is less a missing capability and more a misdirected ambition.

How important is AI readiness when it comes to attracting investment, talent and high-growth businesses to the UK?

It’s become increasingly important. Investment and high-growth businesses gravitate to environments that support modern technology architectures, innovation and rapid execution. True AI readiness is not just about having access to models; it is about the data foundations, governance and operational capabilities to turn those models into measurable outcomes. Showing that an ecosystem supports speed, auditability and data control will matter more and more for attracting investment and talent.

If you could recommend one priority for policymakers looking to strengthen the UK’s AI position over the next five years, what would it be?

I’d spend it on accountability for autonomous action and resist the temptation to write one big AI Bill to deliver it. A single all-encompassing Act tries to pin down a moving target and ages badly as the technology shifts. Regulate where the risk actually lands, sector by sector, and anchor it to one durable principle: when an AI agent or model takes an action in the real world – moves money, reschedules an aircraft, changes a maintenance plan – accountability is clear and auditable in the same way it is for employees. Firms build in the oversight, the logging and the ability to roll back, because they carry the consequences. It’s also technology-neutral, so it survives the next model generation and gives industry the confidence to move agents out of the sandbox and into operations that matter.

It also keeps humans in frame, which is the part that risks getting lost. Accountability can never sit with the software; it sits with the organisation that deploys it and benefits from it. Get that one principle right and you’ll have done more for safe adoption than an impossibly broad framework.