Trustworthy AI series, part one: why trust accelerates operational AI

AI is advancing, and organisations across the UK are experimenting with forecasting, optimisation, and decision-support systems at scale. At the same time, regulations are tightening and scrutiny of automated decision-making is growing. The European Union Artificial Intelligence Act is coming into full force. The question is no longer if AI systems can be built, but if they can be trusted to operate in high-consequence, real-world environments. 

In sectors such as energy, transport, defence, and critical public services, deployment success is rarely determined by novelty or model accuracy alone. What makes or breaks adoption is responsibility. Leaders are increasingly accountable for AI delivering value beyond the pilot stage. They’re responsible for how it behaves when it influences live operations and safety margins. 

As mathematical and operational research specialists, who have supported critical national infrastructure through decades of technological change, Smith Institute brings a distinct perspective on this challenge. We know how trust is built. It requires rigorous, measurable foundations that allow complex systems to be understood and governed over time. 

This first Insight sets the stage for the series by examining why trust has become the gating factor in AI deployment, and why it now acts as an operational accelerator rather than a hindrance. 

Artificial intelligence is no longer confined to innovation teams or proof-of-concept projects. It is moving into core operational functions, such as energy system balancing and infrastructure planning, logistics, maintenance, and risk assessment. As this shift accelerates, accountability moves with it. 

Analysis by the Office for National Statistics shows that only a small proportion of UK firms currently report using AI in their methods or processes. This is not a reflection of weak technical capability, as many organisations demonstrate impressive performance in pilot environments. The challenge comes when AI is exposed to live operations where safety, compliance, and resilience are non-negotiable. 

Industry research reinforces this picture. Only around one in ten AI pilots reach full production environments. Many of these initiatives stall because organisations cannot clearly demonstrate how risk is controlled, how model behaviour can be explained under pressure, or how accountability is exercised when decisions depend on machine-generated outputs. 

The gap between proof-of-concept and business-as-usual deployment isn’t down to technical capability. It's about trust. 

Building operational trust requires deliberate design choices from the outset. 

Impact assessments help surface systemic dependencies and failure modes before they arise in live operations. Evidence packs capture assumptions, limitations, and uncertainty in a way that supports audit and senior decision-making. Governance structures define ownership, escalation pathways, and decision rights, which become vital as systems interact with safety-critical processes. Alignment with recognised standards provides a shared reference point across technical, risk and leadership teams. 

Taken together, these elements demonstrate trust and allow organisations to showcase more than system performance. They deliver evidence that systems can be trusted to continue working as conditions change. 

Strategic and operational decision-makers are responsible for safe, compliant, and scalable deployment. For them, these elements reduce avoidable friction, shorten the transition from pilot to production, and increase the likelihood that investment delivers sustained operational value.