Most organisations seem to believe they are losing the race to harness the full potential of AI. Very few have stopped to ask whether they are measuring themselves against the right finish line.
In reality, most organisations are working through the same range of issues hindering AI adoption – they just aren’t shouting about them from the rooftops for fear they will be seen as laggards rather than thinking carefully before acting. Leadership teams across very different sectors tend to assume competitors are further ahead than they are themselves.
That perception gap matters because it shapes behaviour. Organisations that believe they are already late often rush towards visible activity and mistake that activity for progress.
Recent reporting on “tokenmaxxing”, where employees deliberately increase their use of generative AI tools, shows why organisations need to be careful about what they count. Higher usage may look impressive on a dashboard, but it does not automatically mean work is improving, decisions are better or productivity is increasing.
Adoption figures do not tell the full story
Government research suggests around one in six UK businesses currently use at least one AI technology. The figure captures formal adoption, though it says far less about the informal and uneven ways AI is already entering day-to-day work.
A large proportion of employees are already using AI-enabled tools embedded within software they rely on every day. Others are experimenting informally with generative AI to summarise information, speed up administration or support research and drafting tasks.
Elsewhere, businesses have already run meaningful pilots and identified where AI could improve specific workflows, though they are being selective about what to scale further.
Any caution about AI use is often interpreted externally as hesitation when it is usually a sign that organisations are trying to avoid deploying tools without a clear operational purpose or more certainty about the through-life costs.
A business can generate a great deal of AI activity without materially changing outcomes. Employees may test multiple tools, and leadership teams may announce new initiatives, while the underlying workflow remains largely unchanged once the initial attention fades.
The model race distracts from the harder work
Every few months, a new model release dominates the conversation around enterprise AI. Benchmarks are compared, performance gains are dissected and organisations are encouraged to believe the latest release materially changes what is possible for them operationally. For most organisations, it rarely does.
The harder problems are usually operational and cultural rather than technical. An organisation with inconsistent data, weak governance and unclear ownership structures will not solve those problems by upgrading to a newer model. The output may improve marginally, though the underlying operational weaknesses remain exactly where they were before.
This becomes more obvious once AI adoption spreads beyond technical teams. Legal, HR, operations and customer support functions are not evaluating model performance through technical benchmarks. They need to evaluate whether the system fits the reality of the work, whether outputs are reliable enough to trust and whether accountability remains clear once AI becomes part of the process.
A more subtle issue is also emerging through the way organisations are being introduced to AI in the first place. New AWS research found that 49% of UK organisations cite a lack of AI and digital skills as their biggest barrier to scaling AI so ‘AI literacy’ has become an immediate focus for many.
However, much of the free training available to businesses comes from providers with a commercial interest in promoting a particular category of AI product (and one with a footprint on the planet many are still trying to assess).
As a result, many organisations receive a narrow view of what AI adoption should look like before they have fully defined the problem they are trying to solve.
What organisations mistake for AI progress
AI readiness is often discussed as though it begins once a tool has been deployed. In reality, most of it depends on the organisation surrounding the technology rather than the technology itself.
Data still sits underneath nearly all of it. Organisations continue to encounter familiar problems around inconsistent data, fragmented systems, weak governance and unclear ownership. More advanced models do not remove those issues because they still depend on the quality of the information feeding them.
The organisations making steadier progress with AI tend to be unusually specific about where they expect value to appear. There is generally a defined process, a measurable operational problem and a clear owner responsible for evaluating whether anything improved once the system was introduced.
Without that clarity, pilots often succeed technically while failing operationally. The demonstration works and enthusiasm remains high for a short period of time, though the deployment never properly changes how work happens once it reaches live environments.
The organisations that learn fastest are usually the ones prepared to evaluate those outcomes honestly. Sometimes an employee who struggles with a tool is not resisting change. They may simply be identifying that the system does not fit the pace, nuance or complexity of the work they actually do day to day.
Foundations matter more than the pace
The organisations that will scale AI successfully may not be the ones generating the most visible activity today. More often, they will be the businesses building the operational foundations carefully enough that AI continues working once the excitement surrounding deployment fades.
A large amount of that work remains relatively invisible externally. Improving data quality, redesigning workflows, clarifying ownership structures and building governance frameworks rarely look like major AI transformation milestones from the outside, though those decisions usually determine whether adoption becomes sustainable once AI moves beyond isolated pilots.
Some organisations are closer than they think because the harder part of AI adoption is rarely the visible part. The work that determines whether AI becomes durable usually happens much earlier inside the organisation’s data, workflows, governance and operational culture.
