If AI is supposed to transform the whole enterprise, it should be owned by the people who run it. So why isn't it?
AI Used to Be an IT Problem
Not long ago, AI in the enterprise was genuinely an IT problem. Which vendor? Which model? Which data controls? How do we prevent sensitive information from leaving the perimeter? Those were the right questions, and IT was the perfect team to answer them.
But something changed when AI left pilot mode and broadened its scope. The scarce resource is increasingly transitioning from being access to AI to something harder to measure: the disciplined allocation of intelligence to the work, roles and workflows where it actually creates value.
That is not an IT question. It is a workforce question, a finance question, a risk question and a transformation question. In most enterprises it is going entirely unanswered, because IT is still answering the old questions that are fading in relevance.
Relatively few organisations have reached the most mature state, where AI value reporting actively shapes strategy at the board level. Most fall earlier on the curve, measuring value through shallower metrics and missing out on the true transformational value of AI.
The Token Bill Is the First Signal of Change
One of the clearest signs that AI has outgrown IT governance is what is happening to enterprise token spend.
On the surface, the trend suggests AI is getting cheaper. Analysis of 2.4 billion enterprise API calls shows the blended cost of AI fell 67% from last year. But Jevons Paradox points to the opposite enterprise outcome: as the unit cost of AI falls, organisations use it more often, in more workflows, and at greater scale, causing total AI consumption and spend to rise.
The FinOps Foundation's 2026 State of FinOps report found that 73% of enterprises reported their AI costs exceeded original projections. Price and invoice are moving in opposite directions.
This is not a technology failure. It is a governance failure. As every workflow in the organisation begins to call a model for reporting, summarisation, code review, client communication and internal search, volume expands in ways that per-seat budget models were never built to anticipate. The bill is no longer predictable from the procurement conversation.
Average monthly token spend across enterprise customers has increased 13x since January 2025. The question AI now poses to finance is not "what did we pay per licence?" It is "are we spending intelligence on the right work?" That is a fundamentally different question, and it has evolved well beyond the initial AI rollout.
The Three Risks IT Governance Was Never Built to See
When token spend becomes a line item that CFOs cannot explain and boards cannot contextualise, three distinct risks compound unnoticed.
Cost risk
Provider subsidies and early enterprise pricing have partially obscured the true run-rate of AI consumption. The subsidy model offers generous AI capabilities at consumer-friendly prices to drive adoption, and it has been effective at building user bases. The question is how long it can continue before pricing must reflect actual compute costs. When that correction arrives, organisations without a consumption baseline will have no foundation for negotiation, no ability to identify waste and no data to defend budget decisions.
Operational risk
In the absence of deliberate routing, teams default to the most capable model they have access to for every task. A simple internal summary gets routed through a frontier reasoning model. A document classification task that could run on a fraction of the cost runs on the most expensive infrastructure available. The cheapest production models in 2026 cost around $0.04 per million tokens. The most expensive frontier reasoning models cost upward of $180 per million tokens. That is a 4,500x pricing spread. Model selection should be intelligent at the infrastructure layer. Right now it is neither. The same task sent to the wrong model can be orders of magnitude more expensive than necessary, and this is happening silently, at scale, across every team that has AI access.
Transformation risk
AI applied to a broken or inefficient workflow does not fix the workflow. It accelerates the local task while leaving the end-to-end process unchanged. A step that is thirty percent faster means nothing if the queue behind it is unchanged. The process is no faster. The output is no better. The investment is consumed without creating transformation. Without workflow-level visibility, there is no way to distinguish genuine process change from local optimisation that flatters individual productivity numbers while leaving organisational throughput untouched.
Each of these risks sits in a different function's mandate. What connects them is that none of them are currently visible from an IT deployment dashboard.
Why the Handover Hasn't Happened
The C-suite wants to own this. That much is clear. A 2026 IBM survey of 2,000 CEOs found that 83% say AI success depends more on people's adoption than on the technology itself. The CHRO Association's 2026 survey found that 91% of CHROs rank AI and digitalisation of the workplace as their single top concern, ahead of governance, engagement and talent combined. CFOs across every major industry are calling 2026 the year AI moves from experimentation to proven enterprise-wide impact.
The appetite for ownership exists. The problem is structural: the executives being asked to make workforce, finance and risk decisions about AI are not receiving the data those decisions require. The practical framework for this transfer does not yet exist.
Most enterprises today have AI organised as a technology initiative, with siloed applications across departments. Each function is being held accountable for outcomes it has no instrument to measure. Only 12% of enterprises currently have mature AI governance processes in place, despite agentic AI moving into production at scale across most large organisations. The gap is the absence of the data layer that would connect AI activity to the decisions these executives are being asked to make.
What Specifically Needs to Change
The transition from IT-owned rollout to C-suite-owned outcomes is a data infrastructure decision with three concrete components.
The unit of measurement has to change.
Seats and token volumes are infrastructure metrics. They were designed to manage access, not value. The metrics that serve this moment are adoption quality by role and workflow, model routing efficiency, data class exposure by use case, and the signal that separates genuine work from activity. The useful measure is not price per million tokens. It is tokens per useful work outcome. That single reframe shifts the conversation from procurement to performance.
The same telemetry needs to serve all four functions.
The CFO sees cost. The CIO sees architecture. The CRO sees risk. The CHRO sees workforce change. Right now those four conversations are happening in parallel with different data, different dashboards and no shared foundation. The most successful AI transformations in 2026 are co-led by executives who share accountability for both technology outcomes and business value creation. Siloed ownership consistently leads to underperformance. Role-specific visibility built on a shared data layer is what closes the gap.
Workflow-level visibility has to replace provider-level reporting.
Provider dashboards do not show where AI was consumed, by whom, on what class of work, or whether the output was worth keeping. An organisation running AI across KYC processes, internal document review, engineering workflows and client communication cannot make intelligent routing, training or governance decisions without knowing which of those workflows is high-value, which is high-risk, which is producing good output and which is quietly burning budget on work that a cheaper model could handle. That intelligence does not come from Microsoft, OpenAI or any provider. It comes from a layer that sits above them and maps consumption to context.
Our Solution
The enterprises closing this gap are not doing anything structurally complicated. They treat AI differently, with a management view designed to improve every aspect of AI and increase its impact.
That is what the AIM Score is built to be: a single, board-ready metric that reflects your organisation's AI maturity. It blends prompt quality, cost efficiency, compliance, workflow adoption, and usage patterns into a 0–100 score that updates as your organisation evolves.
PromptLeash helps you measure how effectively your organisation is using AI today and builds the foundation for automated optimisation and governance tomorrow, via intelligent recommendations that show where the organisation stands, how it compares to peers, and what to do next.