Name in original language
BolognAI - Intelligenza Urbana
Initiative overview
The Municipality of Bologna has framed responsible AI governance as a system of operational safeguards, rather than as a stand-alone pilot project. The objective is to balance speed of experimentation with robust control: before implementation, every AI initiative must be underpinned by adequate data, explicit accountability, the appropriate level of involvement, a risk assessment, and safeguards for citizens and employees.
To avoid an approach that is either too slow (top-down) or too uneven (bottom-up), a hybrid model has been adopted (centralised and Hub & Spoke). A central co-ordination function defines shared standards and common approval gates (prioritisation criteria, risk levels, minimum requirements for data/security/transparency), while the identification and structured analysis of needs remain the responsibility of operational units to ensure alignment with day-to-day processes. Strategic oversight and final decision-making are ensured by the Integrated Directorate, which safeguards overall coherence.
Key roles and functions are defined in the PDF file “organisational model roles and responsibilities ENG”: AI Ambassadors; Transformation Demand; AI Transformation Room; and BIS (Bologna Innovation Square).
In addition, assets such as the Civic Digital Twin strengthen synergies for AI use in areas such as simulation and decision support, providing infrastructure and methodologies for data governance and for addressing ethical and regulatory challenges (e.g., a technical–regulatory sandbox).
The model is designed to mature over time. Following an initial survey of around 4,000 people to map AI knowledge and perceptions, core governance capabilities (risk and impact assessment, use-case portfolio management) are expected to be progressively internalised. At the outset, these capabilities can be activated externally (via the AI Transformation Room) to accelerate delivery. As organisational maturity advances, a gradual decentralisation of responsibilities is envisaged.
The operational core is an end-to-end process covering the lifecycle of AI initiatives (with a RACI and standardised deliverables): Intake and validation: request form, expected KPIs, data mapping, initial risk classification and assessment of alternatives (reuse, build/buy); Controlled experimentation: PoC with acceptance criteria, quality/robustness testing, checks on security/privacy/bias, and definition of human oversight; Go-live and scaling: go-live checklist, model/supplier register, continuous monitoring of performance and drift, incident logging, and periodic reviews.
The demand phase is complemented by engagement actions (targeted training, guidelines, AI Ambassador communities) to embed compliance and responsibility directly into behaviours, making risk and data-assessment safeguards stable requirements integrated into the organisation’s ordinary pathways.
