TL;DR

Ask a UK executive whether their organisation can automatically work out why an AI workload failed, and 59% say yes. Ask the infrastructure engineers who actually receive those alerts, and the figure drops to 34%. That 25-point divergence, from a Virtana survey of 238 UK decision-makers, is wider than the 17 points recorded in the United States.

Running what you cannot see

The headline finding is blunter than the confidence gap: more than half of organisations — 53% — admit some of their production AI runs without proper monitoring.

UK enterprises are also scaling faster than American ones — 59% across teams versus 54% — while reporting less stable operations. Only 26% called AI workload performance highly predictable, against 34% in the US. Deploying faster than you can explain what your systems do is a specific combination, and this is what it looks like in numbers.

At 75% of UK enterprises, an automated alert is what fires first when a workload fails. Detecting is not diagnosing, though. Some 8% need several teams working in concert, which can stretch to hours or days; 12% correlate by hand between tools; 32% can see one domain only; and just 47% get to root cause automatically in every infrastructure domain.

What is being dropped to pay for it

The trade-offs are the part UK boards should read twice. Cost-optimisation work is being deprioritised by 54% of respondents, legacy modernisation deferred by 48%, training and upskilling deprioritised by 43% — and security and compliance reviews by 39%.

That last figure sits awkwardly beside the regulatory position. Virtana chief executive Paul Appleby argued that regulatory accountability and observability have stopped being separable concerns in Britain: the systems proving an AI deployment works properly are the same ones that satisfy a regulator. Cutting compliance review while expanding AI use moves in the opposite direction to UK GDPR obligations and sector oversight in financial services and healthcare.

Hardware costs are shaping this. Two-thirds — 66% — said premium AI hardware prices have changed how they invest, pushing work into hybrid environments and consolidating systems mid-flight.

Looking forward

Final say on AI investment rests with IT leadership at 73% of the organisations surveyed, and those same leaders report the highest confidence in diagnosing failures. The people signing off are the people furthest from the alert.

This tracks with our coverage yesterday of Moody’s warning that banks depend on AI systems they cannot properly test. Two separate pieces of research, a day apart, describing the same structural problem: assurance has not kept pace with deployment, and the gap is widest exactly where the authority sits.