UK execs overconfident on AI failure diagnosis, survey finds

Virtana has published UK research into failures in enterprise AI infrastructure, highlighting a wide gap between executive confidence and engineers' views on diagnosing those failures.

Based on responses from 238 UK enterprise decision-makers, the survey found that 53% of organisations are running AI infrastructure they cannot fully observe. It also found that 59% of UK executives believe their organisation can automatically identify the root cause of an AI workload failure across systems, while only 34% of infrastructure engineers who handle those alerts agree.

The 25-point gap is larger than in the United States, where the equivalent divergence was 17 points. At the same time, UK enterprises appear to be scaling AI slightly faster than their US counterparts, with 59% doing so across teams compared with 54% in the US.

That faster deployment is not matched by predictable operations. Only 26% of UK enterprises described AI workload performance as highly predictable, compared with 34% in the US.

Operational strain

The findings point to a broader tension inside large organisations, as boards and senior leaders push AI investment while technical teams manage the consequences when systems fail. Automated alerting is now the first response to AI workload failures for 75% of UK enterprises, yet many still struggle to move from detecting an issue to explaining it.

Just 47% of respondents said they can identify root cause automatically across all infrastructure domains. Another 32% said they can only see a single domain, 12% require manual correlation across tools, and 8% need multi-team coordination that can take hours or days.

Among the toughest monitoring challenges were cost and efficiency metrics, data pipeline visibility, storage and throughput, network bottleneck detection, and GPU utilisation tracking. The figures suggest many companies have reached production deployment before resolving basic visibility problems across the systems supporting AI workloads.

Governance pressure

The research also suggests some organisations are setting aside supporting work as AI demands rise. In the UK, 54% said they are deprioritising cost-optimisation initiatives, 48% are deferring legacy infrastructure modernisation, 43% are deprioritising team training and upskilling, and 39% are deprioritising security and compliance reviews.

This matters because UK companies face tighter scrutiny over data handling and AI oversight than many peers elsewhere. The findings suggest the governance challenge becomes more acute as businesses expand AI use while reducing work tied to compliance and risk controls.

Cost pressures also appear to be shaping those trade-offs. Some 66% of UK enterprises said the cost of premium AI hardware has changed their investment approach, prompting them to shift workloads across hybrid environments and consolidate systems while AI operations continue to run.

Final authority over AI investment sits with IT leadership in 73% of the UK organisations surveyed. Those same senior teams also report the greatest confidence in their ability to diagnose failures, underscoring the gap between decision-makers and engineers.

Paul Appleby, Chief Executive Officer at Virtana, linked the issue directly to accountability. "In the UK, operational observability and regulatory accountability have become inseparable concerns. Enterprises are deploying AI into production under UK GDPR, emerging AI Act obligations, and sector-specific oversight across financial services, healthcare, and national infrastructure. The systems required to prove an AI factory is performing properly are the same systems that satisfy a regulator. For the boards responsible for AI oversight, sovereignty ultimately comes down to whether an organisation can see, attribute, and prove what its AI systems are doing across every environment it runs in," he said.

Scaling challenge

The survey indicates that UK enterprise AI use has moved beyond experimentation. Alongside the 59% already scaling AI across teams, a further 17% said they are running early production workloads. The share scaling AI was highest among organisations with revenue between USD $1 billion and USD $3 billion, at 70%.

Respondents said the two most important steps to improve their ability to scale were a unified platform with visibility and control across AI and infrastructure layers, and AI-driven root cause analysis without manual correlation. Clearer returns from existing AI spending also ranked among the leading conditions for further expansion.

Appleby said the UK findings reflect the structural problems seen in the US at a later stage of deployment. "The US data showed enterprises scaling AI faster than their ability to govern it. The UK data shows that same structural challenge at a more advanced stage of deployment and within a more demanding regulatory environment. UK executives are authorizing infrastructure investment based on a level of confidence in operational readiness that their own engineering teams do not share, and that gap is wider here than in the US. As AI workloads move into production, that gap becomes an operational and governance risk. When compliance reviews are being reduced as AI regulation expands, organizations lose the visibility and control they need to manage performance, cost, and risk at scale," he said.