Why AI first looks like an IT programme
When a company launches an AI programme, the reflex is familiar. It picks a tool, negotiates licences, activates accounts and trains a few teams. It is the classic path of a software rollout.
The reflex is understandable. A tool like Microsoft 365 Copilot arrives through IT, which controls the environment, security and timeline. And results can be measured quickly: a number of licences, an activation rate.
In many cases, these figures reassure more than they inform. An activated licence says nothing about what has changed in the work.
What this view overlooks
Installing a tool does not automatically transform work. Traditional software replaces one way of doing things with another. Generative AI opens a field of uses: each person applies it to different tasks, with variable results. It is up to the organisation to decide where it belongs.
Recent studies point to a gap between ambition and action. In the Roland Berger study of 472 executives, 62% of respondents anticipate a major or radical transformation of their operating model because of AI. Only 38% have started to act. In addition, 59% believe their leadership is not sufficiently prepared.
These are reported results, from a specific sample, with no universal reach. But they describe a climate that many executive committees will recognise: the conviction that something is changing, and the difficulty of knowing where to start.
AI transforms the work system
AI affects processes, decisions, roles, skills and performance indicators. Take a generic example: a legal department using AI to analyse contracts. The gain does not come from the tool. It comes from how review is split, what is still checked by hand, and who answers for the result.
Microsoft's 2026 Work Trend Index points the same way. It is based in particular on a survey of 20,000 professional AI users in ten countries. According to its results, 19% are in a situation where individual capabilities and organisational readiness reinforce each other. 31% experience a misalignment.
I read this finding with caution: it is a vendor survey, not proof of cause and effect. But it is a reminder that an employee comfortable with AI, in an organisation that has planned nothing, can end up working alone in a corner.
The trap of a technology-only programme
A programme driven only by IT risks producing licences and pilots, but little real usage. This is not a criticism of IT. It does what it is asked to do: deploy, secure, support.
The problem is that nobody else takes ownership of the subject. The business waits for the tool to arrive. Legal is consulted late, to sign off. Human resources discover that roles and skills are shifting. Pilots multiply, with no criteria for choosing which ones to scale.
Responsibilities to share differently
The business must define the expected outcomes and the situations where AI brings concrete value. Nobody else can do it. A customer service team or a finance department knows where it loses time, where quality varies, where a decision drags.
IT, security, legal, human resources and the business must build the conditions for adoption together. Each brings a piece:
- IT provides the platform, access and integration
- Security and compliance set what is allowed, and on which data
- Legal clarifies responsibilities and risks
- Human resources handle skills, roles and support
- The business owns the expected outcomes
The World Economic Forum, drawing on contributions from more than 450 executives, identifies five conditions for scaling: human accountability, operating model redesign, skills development, trust through transparency and disciplined experimentation.
How to move from deployment to real usage
I suggest four moves, simple to state and demanding to sustain:
- Choose a few work situations, each with a business sponsor who states the outcome expected
- Agree the rules together: permitted data, human validation, accountability
- Treat each pilot as an experiment, with a question, a duration and a criterion to scale or stop
- Track change in the work, not only the tool's activity
The last move is decisive. Success must be measured by the changes achieved in the work, not only by the number of users activated. Depending on the use case, this can be the time to process a file, the quality of a first draft or the time freed for higher-value tasks.
The question executives should now ask
Instead of « how many people have access to the tool? », I suggest asking: what has changed in the way we work, and who answers for it?
If the answer is vague, the problem is probably not technical. It is a question of decisions, roles and measurement. It gets settled at leadership level, before adding more licences or launching another pilot.
Do you have an AI programme moving more slowly than planned? Let’s talk, to identify what is blocking it, realign stakeholders and get adoption moving again.
Sources
- The AI-First Organization, Roland Berger, 3 July 2026
- 2026 Work Trend Index Annual Report: Agents, human agency, and the opportunity for every organization, Microsoft, 5 May 2026
- Organizational Transformation in the Age of AI: How Organizations Maximize AI's Potential, World Economic Forum, in collaboration with Accenture, 16 March 2026