The AI utilisation gap is the distance between the AI licences an organisation has paid for and the work those licences actually change. Access is bought in a single procurement decision. Utilisation has to be built, role by role, and most organisations have stopped at access. The result is a growing licence bill with little movement in cost, speed or quality.
The problem goes by several names: shelfware, the access-to-adoption gap, low AI utilisation, licence underuse and the “search engine trap”, in which people use an AI assistant to look things up rather than to do part of the job. Each describes high availability, shallow use and a return the CFO cannot find.
Utilisation is not only a commercial question. Article 4 of the EU AI Act (Regulation (EU) 2024/1689), applicable since 2 February 2025, requires providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff. ISO/IEC 42001:2023 requires organisations to determine the competence needed by people doing work that affects AI performance and to ensure they have it through education, training or experience. A licence without competence meets neither expectation.
The data explains why the return is missing. More than three-quarters of leaders and managers use generative AI several times a week, but regular use among frontline employees has stalled at 51% (Source: BCG, AI at Work 2025: Momentum Builds, but Gaps Remain, 2025). The same research finds that the share of employees who feel positive about generative AI rises from 15% to 55% when they have strong leadership support (Source: BCG, AI at Work 2025: Momentum Builds, but Gaps Remain, 2025).
This analysis draws on a recent conversation with Marco Galimberti, CEO of Venchi Asia Pacific, about his experience leading a global premium brand across manufacturing, retail and consumer engagement.
Why do AI licences fail to produce a return?
AI licences fail to produce a return because access was treated as the outcome. As the discussion put it plainly, most organisations “gave everyone access to AI” and then watched people use it as a search engine. A search engine answers a question. A tool changes how a piece of work is done. Only the second changes the cost base.
Galimberti’s argument is that the real shift is in use, not access. When a licence is rolled out without a view of which work it should change, people default to the most familiar pattern: ask a question, read an answer, carry on as before. It may save minutes, but it will not show up in a margin.
Where does AI create measurable value first?
AI creates measurable value first in repeatable operational work, where the output can be checked and the saving counted. Galimberti points to process automation, pattern recognition and shift optimisation as the areas of immediate return. The longer-term opportunity is different in kind: using AI to understand and respond to what customers actually want. For the full range of value levers, see Is AI ROI Only About Job Cuts?
For a CFO, the distinction matters because the two horizons need different business cases. Near-term use cases should carry a measurable saving. Longer-term ones are strategic bets and should be funded and judged as such. Mixing them leaves licence spend defended by anecdote. We set out the wider distinction in Automation or Augmentation?
Why does experience matter more than enthusiasm?
Experience matters because AI output has value only when someone can tell whether it is right. Galimberti’s view is that AI currently works better in the hands of experienced professionals who can verify what it produces. He also warns that an overabundance of data is more dangerous than a shortage of it: more output is not more insight if nobody can judge it.
Spreading licences evenly assumes every role benefits equally. It does not. The highest early return usually sits with experienced people in defined processes.
The AIPEX view: fund utilisation, not seats
At AIPEX we treat an AI licence as a commitment to change specific work, not as a software purchase. The executive decision is to stop approving seats and start approving use cases, each with an owner and a measure. Three changes make that practical:
- A utilisation register. For every licence tranche, record which roles hold it, which workflows it is meant to change and what measure will show the change. Renewal depends on the register, not on login counts.
- Competence before scale. Fund training and coaching as part of the licence cost, and target it at the roles where value is expected first. This is also the most direct way to evidence the AI literacy and competence the frameworks above expect.
- Visible executive use. Galimberti argues that leaders who use AI openly and normalise it across their teams are the ones driving real adoption. The BCG data above supports him. We explore this further in Leaders Prompt First.
| Measure | Access mindset | Utilisation mindset |
|---|---|---|
| Unit of investment | Seats purchased | Workflows changed |
| Success metric | Logins and active users | Time, cost or quality moved in a named process |
| Training | Optional, generic | Budgeted, role-specific, measured |
| Renewal decision | Automatic | Conditional on evidence of use |
How should the CFO measure licence ROI, and what must the executive team change to improve it?
The CFO should measure licence ROI against the workflows each licence was meant to change, not against usage statistics. Active users show only that people logged in. The measure has to sit in the process itself: cycle time, rework, cost per transaction or hours released and redeployed.
The executive team, in turn, must change three habits: approving licences without named use cases, treating training as optional, and assuming adoption continues once the launch is over. We cover how to keep it going in From Hype to Habit.
What should the board ask about AI licence spend?
The board should ask what the licences have changed, not how many people have them:
- What do we spend each year on AI licences, and which named workflows is that spend meant to change?
- For each of those workflows, what measure has moved since the licences were issued?
- What proportion of licence holders have received role-specific training, and how do we know it worked?
- Which licences would we not renew today if renewal depended on evidence of use?
- How do we demonstrate the AI literacy and competence our regulators and standards now expect?
If nobody at executive level can answer these, the gap is ownership, not technology. A Fractional Chief AI Officer can own the utilisation register and hold each licence tranche to a measurable return. The Executive Strategy Day maps your current AI spend to the workflows it should be changing and identifies where the return is being lost.
Frequently asked questions
Why are staff using AI tools like a search engine?
Because licences were rolled out without a view of which work they should change, so people default to asking questions and reading answers. That use saves minutes but rarely changes cost, speed or quality in a measurable process. For a full explanation, see You Bought the AI Licences. Where Is the Return? →
Is logins or active users a good measure of AI licence value?
No. Active users show only that people logged in, not that a process became cheaper, faster or more accurate. Measure licence value against the named workflows each tranche was meant to change, such as cycle time, rework or cost per transaction. For a full explanation, see You Bought the AI Licences. Where Is the Return? →
How can executives get more employees to use AI regularly?
Use AI visibly themselves and fund role-specific training as part of the licence cost. BCG found that the share of employees who feel positive about generative AI rises from 15% to 55% with strong leadership support. For a full explanation, see Leaders Prompt First: Why AI Adoption Starts in the Boardroom →