New research has revealed a significant AI governance gap with just over a third of leaders saying AI-generated outputs are consistently reviewed – despite security and compliance concerns being the biggest cited blocker to their AI programmes delivering greater value.
The research – based on a survey of senior leaders at medium-to-large UK organisations by digital agency Reading Room – found that while 83% of organisations have an AI governance framework in place, this often isn’t translating to day-to-day practice. Only 38% of leaders said AI-generated outputs are ‘always’ reviewed, while 7% said they are ‘rarely’ or ‘almost never’ reviewed, leaving the majority reviewing outputs inconsistently.
A third of respondents named legal and compliance teams as the greatest internal source of skepticism about AI, however, that concern doesn’t appear to be translating into stronger review practices for most.
Off the back of this, Reading Room experts are urging organisations to focus on building a ‘review muscle’ – applying routine, systematic scrutiny of AI outputs to check for bias, errors, or compliance breaches – rather than defaulting to trust.
Amanda Falshaw, AI Enablement Lead at Reading Room, commented: “Organisations are worried enough about AI risk to call it their biggest blocker to AI having a bigger impact, however, we’re seeing a direct contradiction in the fact that so many still aren’t reviewing outputs. Organisations must treat human oversight as non-negotiable, as even the strongest governance policy can fail if consistent human oversight isn’t built in to review AI outputs. At a very basic level, we should all be asking of AI outputs: What’s missing? What assumptions have been made? How would we verify this? Would I be comfortable putting my name to this?”
The research also revealed mixed views on AI’s effectiveness. While most (97%) respondents said AI has delivered some level of measurable impact, for almost 1 in 4 its real-life impact has fallen short – with 22% of respondents reporting that its impact was ‘lower’ or ‘significantly lower’ than expected. This is being compounded by a measurement issue, with a lack of effective measurement coming in the top three cited barriers to AI effectiveness.
Most value to date has come from efficiency and productivity gains, rather than top-line growth — just 18% of respondents said they had seen direct financial gains through new or additional revenue.
Amanda continued: “AI absolutely has the potential to generate real impact for organisations, however it heavily depends on the business case around it. Companies should be starting with a business problem, not an AI solution. The strongest programmes begin with a clear organisational challenge or opportunity. They also need to be looking at more than just efficiency gains when it comes to measurement. For example, instead of looking solely at time-tracked or minutes saved, organisations could look at decision or work quality improvements, improved customer experience or innovation capacity.”