Public university (Europe)

Context

A public university had accumulated 23 uncoordinated AI initiatives: licences contracted by individual faculties, isolated workshops, contradictory policies across departments and no impact indicators.

The most revealing symptom was not the dispersion but the asymmetry: while the institution debated whether to regulate AI, its students were already using it daily — and so was part of the faculty — with no criteria, no traceability and no institutional conversation. The real risk was not “AI” but the growing gap between everyday practice and the institutional framework: every day without a strategy was a day in which the strategy was written, de facto, by tool vendors and individual usage.

Macro–meso–micro reading

• Macro: there was no institutional position on the role AI should play in the educational model, research and management. Without that decision, everything else was patchwork.
• Meso: each faculty improvised its own policy — from prohibition to criteria-free enthusiasm — creating comparative grievances among students and confusion among faculty.
• Micro: the teaching conversation was trapped in the wrong question (“how do I detect that AI did this?”) instead of the relevant one (“what must the human keep doing, and how do I make it observable?”).

Intervention

Through BeTalent, an institutional AI strategy process in four phases:

1. Inventory and maturity diagnosis. Mapping of the 23 existing initiatives — which tools, which contracts, which data they touch, who actually uses them — and a maturity diagnosis across five dimensions: strategy, governance, competencies, infrastructure and culture. The inventory surfaced duplicated spending, data-protection risks nobody was overseeing and, also, isolated pockets of excellence the institution didn’t know it had.

2. Governance framework with usage criteria. Not a rulebook of prohibitions but a decision matrix: which activities can be delegated to AI, which require human oversight, and which must be preserved as human activity — because they are precisely where learning happens or where academic responsibility is at stake. The framework defines who decides what (classroom, degree programme, institution), how new tools are evaluated before being contracted, and how all of it is revisited as the technology changes.

3. Role-contextualised literacy. No generic “introduction to AI” courses: distinct itineraries for teaching (redesigning activities and assessment), research (integrity, authorship, peer review) and management (processes, data, decisions), each anchored in that role’s real problems and aligned with the literacy obligation of Article 4 of the European AI Act.

4. Two-year roadmap with indicators. Prioritisation of the surviving initiatives into strategic lines with an owner, budget, metrics and semester review points — designed to survive changes of rectoral team, which is where most university strategies die.

Results and lessons

• Governance framework approved, with criteria for delegation and for preserving human activity: the institution can now answer “can we use this?” in days, not months of committees.
• Literacy programme aligned with Article 4 of the AI Act: the legal obligation became a lever for teaching transformation rather than a compliance formality.
• Reduced dispersion: from 23 initiatives to 6 strategic lines, each with an owner and metrics — and duplicated spending on unused licences released.
• The institutional conversation changed register: from “how do we protect ourselves from AI” to “what do we decide to do with it” — and that difference of question is, in itself, the most durable result.
• Key lesson: governance was accepted when it was presented as enabling (criteria to move forward) rather than restrictive (a list of prohibitions). The most important document was not the one that said “no”, but the one that said “yes, like this”.