Governing Urban AI from the Frontline: A Stage-Gate Framework for Municipal Algorithmic Decision-Making

Highlights

What are the main findings?

• A stage-gate governance framework is proposed to guide municipalities in managing AI-driven algorithmic decision-making across planning, deployment, and oversight stages.
• The study synthesises global AI governance principles and city-level practices to demonstrate that locally grounded, participatory, and adaptive governance is essential for responsible urban AI.

What are the implications of the main findings?

• Local governments should play a leadership role in operationalising responsible AI, supported by structured governance tools such as stage-gate checklists and participatory oversight mechanisms.
• The framework provides practical guidance for aligning urban AI innovation with equity, accountability, and sustainability, while highlighting key research and policy gaps for future municipal AI governance.

Abstract

Artificial intelligence (AI) is increasingly embedded in how cities are governed, shaping decisions on mobility, land use, public services, and environmental management. Yet urban AI is predominantly governed through fragmented frameworks designed at national or corporate scales, offering limited guidance for municipal decision-making and overlooking place-specific social and ecological consequences. As the level of government closest to everyday urban life, cities are uniquely positioned to steer AI toward public value, but face persistent tensions between efficiency, equity, accountability, and sustainability. This paper argues that responsible urban AI cannot be governed through top-down or one-size-fits-all approaches. To address this, the study aims to conceptualise and advance a ground-up model of responsible urban AI governance that places cities and local governments at the centre of decision-making. It addresses the following research question: How can municipal authorities translate high-level ethical principles into practical, context-sensitive governance arrangements that respond to local capacities, risks, and public values? Drawing on global governance principles and illustrative city experiences, we propose a locally grounded, stage-based framework for municipal AI governance. The framework addresses institutional capacity gaps, fragmented responsibilities, and algorithmic externalities, advancing a participatory, place-sensitive, and adaptive model that aligns urban AI innovation with democratic legitimacy, social justice, and sustainable urban futures.

1. Introduction

Artificial intelligence (AI) is no longer a distant or experimental frontier in the public sector; it has become an active force shaping how cities are governed. Across different tiers of government, AI systems are increasingly embedded in service delivery, administrative coordination, and policy implementation, reshaping interactions between institutions, infrastructures, and communities [1]. While AI adoption is often framed as a technical necessity or a pathway to efficiency, its growing presence in urban governance raises deeper questions about public value, democratic legitimacy, and accountability [2].

National and state-led initiatives have played a decisive role in legitimising AI in government by establishing ethical principles, regulatory benchmarks, and innovation priorities. For example, South Australia’s recent investments in AI for healthcare and policing reflect a broader political commitment to institutionalising AI in essential public services [3]. These top–down approaches can enable local governments to experiment with AI applications by providing resources and strategic direction. However, they also constrain municipal discretion, limiting the ability of cities to tailor AI governance to place-specific needs and values [4]. This positions cities at the frontline of AI implementation while leaving them peripheral in the design of AI governance frameworks.

A persistent gap exists in how widely cited frameworks, such as the OECD AI Principles, the UNESCO Recommendation on AI Ethics, and the EU AI Act, are enacted at the municipal level. Although these frameworks articulate important global norms, they provide limited guidance on how local governments can translate high-level principles into actionable governance practices under real-world constraints [5]. This gap highlights the need for governance approaches that are sensitive to local institutional capacities and contextual realities.

This paper aims to conceptualise and advance a ground-up model of responsible urban AI governance that places cities and local governments at the centre of decision-making. It seeks to address the research question: How can municipal authorities translate high-level ethical principles into practical, context-sensitive governance arrangements that respond to local capacities, risks, and public values? By synthesising insights from urban governance, public administration, and AI ethics, the paper develops a framework that emphasises participation, institutional learning, and adaptive oversight as core components of responsible AI deployment. In doing so, it demonstrates how cities can move beyond compliance-oriented or technology-driven approaches toward a more reflexive mode of governance that aligns AI innovation with democratic accountability, social justice, and sustainable urban development.

2. Literature Background

Urban AI reflects a broader shift toward data-driven and anticipatory forms of governance, where algorithmic systems increasingly mediate how cities plan, coordinate, and deliver services [6]. This shift moves from automated tasks to autonomous, AI-driven city management, enabling anticipatory governance where AI directly interacts with infrastructure, citizens, and decision-making processes [7,8]. Existing scholarship highlights the potential of AI to support urban efficiency, mobility, environmental management, and service responsiveness, while also drawing attention to risks associated with opacity, bias, surveillance, digital inequality, and uneven social and ecological impacts [9,10,11]. These risks are especially significant in cities, where AI-enabled decisions are embedded in everyday infrastructures, public spaces, and service interactions.

Local governments are already experimenting with AI across domains such as citizen services, transport analytics, planning, compliance, and environmental management [12,13]. However, efficiency gains do not automatically translate into public value. AI systems may privilege institutional or commercial priorities, redistribute power between public agencies and technology providers, or produce exclusionary outcomes if their deployment is not subject to meaningful oversight [14,15]. Public organisations therefore face not only technical and legal challenges, but also deeper governance questions concerning accountability, legitimacy, and the protection of community values [16].

The literature increasingly recognises that responsible AI governance requires more than abstract ethical principles or one-off compliance checks. It requires adaptive arrangements that combine technical safeguards, institutional accountability, public engagement, and continuous learning [17,18]. Governance ecosystems that connect governments, vendors, communities, regulators, and experts can help coordinate AI deployment through shared standards, transparency mechanisms, and accountability processes [19,20]. However, much of this work remains oriented toward national, supranational, or large organisational contexts, with limited attention to how local governments can operationalise these principles under conditions of constrained capacity, fragmented authority, and place-specific public expectations.

This gap is important because municipalities are both implementers and public-facing stewards of urban AI. Their proximity to communities means that failures of transparency, accountability, or fairness can directly affect public trust and democratic legitimacy [21,22,23]. At the same time, local governments often lack the specialised expertise, resources, and formal governance structures available to larger institutions. Context-sensitive governance frameworks are therefore needed to help municipalities translate responsible AI principles into practical decision-making processes that are feasible, auditable, and responsive to local conditions [24,25].

3. Research Design

This study adopts a qualitative, theory-building research design to develop a context-sensitive governance framework for AI in local government. Rather than evaluating specific AI systems or measuring implementation outcomes, the research focuses on conceptual synthesis and framework development, grounded in existing scholarship, policy analysis, and empirically documented municipal practices. This design is well suited to the study’s aim of addressing a recognised gap between high-level AI governance principles and their practical operationalisation in urban contexts.

The analytical stages of the study were conducted using literature, policy documents, and case materials available up to 2023–2024, which define the temporal boundary of the framework development process. While selected more recent sources have been incorporated during manuscript finalisation to reflect ongoing developments and maintain contextual relevance, these sources are used for contextualisation only and do not form part of the analytical basis of the framework. The research design comprises three interrelated and iterative stages (Figure 1), combining systematic literature review, conceptual and policy analysis, and case-informed refinement. Iteration across stages enables continuous refinement of concepts and governance propositions, ensuring both analytical coherence and practical relevance.

Stage 1 Systematic Literature Synthesis: The first stage involves a structured review of academic literature across urban governance, public administration, responsible AI, and digital urbanism. The review focuses on peer-reviewed journal articles, key books, and authoritative review papers addressing AI governance, municipal digital transformation, and ethical technology deployment in cities. A PRISMA-guided screening and selection process was employed to ensure transparency and methodological rigour, with the detailed flow diagram provided in Appendix A. Rather than exhaustively cataloguing studies, the review adopts a thematic synthesis approach, identifying recurring governance challenges, normative tensions, and institutional constraints specific to local governments. This stage yields a set of core governance dimensions (including institutional capacity, ethical oversight, data governance, procurement, and participation) that form the analytical foundation for the framework. The synthesis also reveals systematic blind spots in existing research, particularly the limited attention to municipal-scale decision-making, sequencing of governance actions, and cumulative social and ecological impacts of urban AI systems.

Stage 2 Conceptual and Policy Analysis: Building on the literature synthesis, the second stage translates identified governance challenges into a conceptual framework informed by governance theory and responsible AI scholarship. This stage involves qualitative analysis of major international AI governance instruments (e.g., principles-based and risk-based frameworks) and their implicit assumptions regarding institutional capacity, scale, and implementation. The analysis examines how global AI governance norms are interpreted, adapted, or constrained when applied within local government contexts. Particular attention is given to mismatches between normative expectations and municipal realities, such as fragmented authority, limited expertise, and reliance on external vendors. These insights inform the development of a stage-gate governance logic, which structures AI decision-making into sequenced review points aligned with ethical, organisational, and operational considerations across the AI lifecycle.

Stage 3 Case-Informed Refinement: The third stage enhances the framework’s practical relevance through case-informed refinement, drawing on documented examples of AI use in local governments reported in academic studies, policy reports, and official municipal sources. These cases are used as illustrative reference points to refine and contextualise the framework, rather than as formal tests or comparative evaluations. This stage supports iterative refinement of governance stages, decision thresholds, and institutional roles by comparing conceptual propositions with documented municipal practices. It also highlights persistent implementation gaps, reinforcing the need for adaptive, place-sensitive governance mechanisms rather than uniform prescriptions.

Methodological Scope and Limitations: The research design is intentionally conceptual and exploratory. In addition to peer-reviewed scholarship, the analysis draws selectively on policy documents, and a limited number of preprint sources where these provide current practice-oriented insight. While empirically informed, it does not involve primary data collection or formal case comparison. Accordingly, the proposed framework should be understood as an analytical and operational guide, rather than a validated policy instrument. The design establishes a foundation for future empirical research, including field-based studies, participatory action research with municipalities, and comparative analysis across urban regions. By integrating theory, policy analysis, and practice-oriented insights, this research design balances analytical rigour with urban relevance. It enables the development of a governance framework that is both theoretically grounded and sensitive to the institutional, political, and socio-spatial conditions under which cities govern AI.

4. Analysis

4.1. The Critical Role of AI Governance in Local Government

Local governments occupy a distinctive position within multi-level governance systems because they are responsible for many services directly experienced by communities, including aspects of public safety, public health, spatial planning, and local administration [26,27]. Yet most AI governance debates remain centred on national and supranational institutions, leaving municipalities comparatively underdeveloped as sites of AI governance despite their growing role in AI-enabled service delivery and urban management [28].

The literature identifies five interrelated governance challenges, institutional, ethical, technical, procedural, and participatory, that inform the development of a municipal AI governance framework. Institutional capacity remains uneven across municipalities, especially where limited resources, fragmented authority structures, and constrained technical expertise reduce the ability to govern AI proactively [25,29]. Ethical risks, including bias, privacy infringement, opaque decision-making, and the marginalisation of vulnerable groups, are also difficult to manage where local governments lack formal ethics review mechanisms, public accountability bodies, or operational policy toolkits [30].

Technical infrastructure and data governance present further challenges. Many municipalities face fragmented data systems, weak data stewardship, limited interoperability, gaps in technical expertise, institutional readiness, and underinvestment in cybersecurity and data architecture, reducing the reliability and scalability of AI-enabled services [6,29,31]. Procedural and procurement risks are also significant because local governments often rely on third-party vendors, creating concerns around transparency, accountability, vendor lock-in, and contractual safeguards [28]. Without robust procurement frameworks and governance tools, local governments may adopt systems that are opaque, poorly aligned with democratic values, or inconsistent with international standards such as those promoted by ISO, OECD, and NIST.

Finally, participatory governance remains weakly institutionalised. Although smart city agendas frequently emphasise citizen engagement, meaningful public involvement in AI design, deployment, and review remains limited, which can undermine trust and democratic legitimacy [27]. Participatory AI governance remains fragmented and weakly institutionalised, with lifecycle-wide inclusion and transparency still rare. Together, these challenges show why municipal AI governance requires more than broad principles. It needs coordinated structures that connect technical systems, organisational responsibilities, policy requirements, procurement practices, and public accountability mechanisms.

To address these overlapping challenges, this study proposes a holistic AI governance ecosystem for local governments, illustrated in Figure 2. The ecosystem integrates technical, organisational, regulatory, and ethical dimensions to support responsible, transparent, and inclusive AI use in municipal settings. It functions as a diagnostic and coordination tool, enabling local governments to assess how existing policies, systems, and institutional arrangements align with responsible AI governance principles, while identifying gaps and areas requiring coordination.

Two foundational pillars underpin the municipal AI governance ecosystem: technical infrastructure and structured processes and workflows [32]. The technical layer encompasses AI models, secure data storage, scalable cloud solutions, and robust cybersecurity measures that ensure system integrity and resilience [33]. Real-time data processing is supported through IoT and edge networks, while immersive technologies and geospatial platforms enhance spatial analysis and public engagement [34]. These capabilities are reinforced by clearly defined operational processes, including algorithm updates, data governance protocols, system monitoring, and incident response procedures [29]. Together, these elements promote transparency, reduce risk, and sustain public trust.

AI governance in local government is inherently multi-stakeholder, involving citizens, municipal staff, elected officials, state and federal authorities, technology vendors, technical experts, academics, civil society organisations, and the media [35]. Sustaining this collaborative ecosystem requires sustained investment in capacity building and public education [36]. Transparency initiatives, community engagement mechanisms, feedback loops, and inter-municipal collaboration strengthen accountability and institutional learning [37]. At the same time, staff training, citizen education, and the development of practical toolkits and vendor guidelines equip stakeholders to engage meaningfully with the complexities of responsible AI deployment.

Clear and accountable governance structures are essential for ethical AI use in local government [38]. Strategic oversight is typically exercised by elected councils and executive leadership, while dedicated AI governance committees and independent review panels enhance transparency and public confidence. Audit and risk committees monitor compliance, and community advisory panels ensure participatory input and local relevance [39]. These arrangements must reflect a commitment to ethical and responsible AI, grounded in transparency, accountability, inclusion, and fairness, to ensure that AI systems remain aligned with community values.

Finally, robust policy and regulatory frameworks are necessary to ensure AI systems are trustworthy, secure, and legally compliant [40]. This includes dedicated AI policies, strong data protection measures, and procurement standards that embed ethical and technical safeguards from the outset [41]. Departmental regulations and cybersecurity frameworks provide operational clarity, supported by overarching AI governance guidelines that coordinate policy and manage risk [42]. For coherence and interoperability, local frameworks should align with regional and national strategies and adhere to international standards such as ISO, OECD, and NIST [43].

A continuous feedback mechanism ensures that AI governance remains adaptive to technological change, regulatory developments, and evolving community priorities [44]. Regular audits, citizen feedback channels, and cross-departmental reviews allow municipalities to update technical systems, refine organisational processes, and recalibrate policy standards over time [45]. This reflexive layer prevents governance obsolescence and supports the long-term legitimacy of AI use in local government.

4.2. Governing Urban AI Across Scales: From Global Norms to Municipal Practice

Governance responses to AI have emerged across multiple scales, ranging from municipal mechanisms that manage transparency, risk, and accountability in service delivery, to international and supranational bodies that articulate normative principles and regulatory models for AI oversight. A multi-level governance perspective highlights how these scales interact, with global frameworks establishing shared ethical standards and risk logics that are subsequently interpreted, adapted, and operationalised by local governments within specific administrative, legal, and political contexts. This section examines the relationship between global AI governance architectures and city-level practices to elucidate how responsible AI governance is shaped in practice and to inform the development of the local government AI governance framework proposed later in this paper.

4.2.1. Global AI Governance Frameworks and Municipal Applicability

Global AI governance frameworks provide normative foundations and regulatory reference points that can inform how local governments design, procure, deploy, and oversee AI systems [46]. Although developed primarily at international or supranational levels, these frameworks articulate principles, values, and risk management approaches that municipalities can draw upon when developing context-sensitive governance arrangements.

The OECD AI Principles emphasise human-centred values, transparency, accountability, and system robustness [1]. For local governments acting as AI deployers, these principles translate into practical governance measures such as impact assessments, requirements for explainability in automated decision-making, and the establishment of internal oversight mechanisms within municipal departments. Similarly, UNESCO’s Recommendation on the Ethics of AI advances a rights-based and inclusive normative framework that highlights fairness, human dignity, and social well-being. At the municipal level, these principles can inform procurement criteria, community consultation processes, and safeguards for vulnerable populations affected by AI-enabled public services [46].

The EU’s AI Act introduces a risk-based regulatory approach that categorises AI systems according to their potential societal harm [47]. Although formally directed at member states, its high-risk classification framework is particularly relevant for municipalities deploying AI in sensitive domains such as biometric identification, welfare eligibility assessment, or predictive policing. Local governments can adopt similar risk-tiering logics to determine the intensity of oversight, documentation requirements, and transparency obligations associated with different AI applications [48].

In addition, technical and professional frameworks such as IEEE’s Ethically Aligned Design provide implementation-oriented guidance for embedding ethical considerations into AI system design, procurement contracts, and lifecycle management [49]. These tools assist municipalities in translating abstract global principles into concrete administrative procedures and operational standards [50].

Taken together, global AI governance frameworks do not replace local autonomy. Rather, they function as adaptable reference architectures that provide shared vocabulary, normative benchmarks, and risk categorisation models that municipalities can tailor to their institutional capacities, legal mandates, and community values [51]. By bridging global norms with local implementation, these frameworks support more coherent and context-sensitive AI governance at the municipal level.

4.2.2. Municipal Pathways to Responsible AI Governance

To illustrate how global principles are operationalised in practice, this section adopts a qualitative and illustrative approach, examining city-level innovations in AI governance across diverse local government contexts. The cases are organised using four analytically informed criteria relevant to municipal AI governance: the presence of transparency mechanisms, regulatory stance, participatory and inclusion practices, and sectoral breadth of AI application across planning, public safety, and service delivery.

Cities such as Amsterdam and Helsinki exemplify transparency-oriented governance approaches through the establishment of public algorithm registers. These registers disclose information about local government AI systems, including their purpose, data sources, and decision-making logic, thereby enhancing visibility and accountability [52]. Such initiatives represent emerging institutional responses to algorithmic opacity and contribute to the normalisation of transparency practices within municipal governance.

Barcelona illustrates a participatory and rights-based model of municipal AI governance. As a founding member of the Cities Coalition for Digital Rights and the Global Observatory of Urban AI, the city embeds ethical AI principles within its broader digital governance and smart city strategies [53]. By prioritising citizen participation, data sovereignty, and equitable access to digital infrastructure, Barcelona demonstrates how local governments can align AI development with democratic values and social equity objectives.

Singapore represents a standards-based and innovation-oriented approach to AI governance. Its Model AI Governance Framework, first released in 2019 and updated in 2022, introduced AI Verify, a standardised toolkit that enables organisations to demonstrate alignment with recognised ethical AI principles [54]. Through its National AI Strategy 2.0, Singapore promotes AI for the public good by encouraging responsible AI adoption through voluntary and verifiable governance mechanisms [55]. While operating at a national scale, these initiatives have direct implications for municipal implementation and intergovernmental coordination in urban AI governance.

By contrast, San Francisco exemplifies a precautionary regulatory stance that prioritises civil liberties and rights protection. In 2019, it became the first major city in the United States to ban the use of facial recognition technologies by government agencies [56]. This intervention highlights the capacity of local governments to set ethical boundaries on AI deployment, particularly in high-risk contexts involving surveillance, policing, and civil rights.

Within the Australian context, initiatives led by the Municipal Association of Victoria demonstrate how responsible AI governance principles are translated into sector-specific local government applications. Programs such as Advancing AI Innovation in Local Government, developed in partnership with Greater Dandenong City Council and supported by the Commonwealth Housing Support Program, embed governance structures, ethical oversight, and procurement guidance into statutory planning processes [57]. These initiatives underscore the importance of institutional capacity, transparency, and human oversight in fostering trustworthy AI adoption at the local level.

4.2.3. Identifying the Governance Gap

Taken together, aforementioned cases illustrate a diverse yet uneven landscape of municipal AI governance, spanning transparency-led and participatory approaches as well as standards-based and precautionary regulatory strategies. Although shaped by differing institutional capacities, political contexts, and regulatory philosophies, they show how global principles of responsible AI are selectively interpreted and adapted within local settings. At the same time, they expose the limits of current initiatives, principles, and tools reviewed in Section 3, which remain fragmented across policy domains, scales, and institutional responsibilities and are often poorly aligned with the everyday realities of municipal decision-making. Many existing frameworks are designed for national regulators or large organisations and assume stable capacities, clear mandates, and uniform contexts that rarely apply to local governments, leaving cities to retrofit abstract ethical commitments or technical safeguards into practice without clear guidance on prioritisation, trade-offs, or accountability.

These approaches also tend to underplay the cumulative social, spatial, and ecological impacts of AI as it becomes embedded in urban infrastructures and community life. This persistent gap between aspiration and implementation justifies the need for a new governance framework that is explicitly grounded at the local scale, integrates participation and institutional learning, and is sensitive to place-specific values. Accordingly, Section 4 builds directly on the limitations identified in Section 3 to advance a locally anchored and adaptive framework that translates shared governance principles into structured and actionable decision points across the AI lifecycle in municipal contexts.

4.2.4. Local Government Realities and the Need for Context-Sensitive AI Governance

Effective AI governance at the local level must account for municipal capacity constraints, fragmented responsibilities, and heightened public accountability obligations. Local governments deploy AI across diverse service domains, often with limited technical expertise and uneven institutional readiness. A uniform governance model is therefore unlikely to be practical or desirable. Instead, local AI governance requires adaptable mechanisms that balance ethical commitments with local legal, administrative, and socio-political realities [58,59,60].

Recent scholarship similarly emphasises that high-level AI principles must be supported by operational governance mechanisms. Ref. [61] identify both best practices and persistent barriers in AI governance, highlighting tensions between innovation and risk mitigation. Ref. [44] advance the Hourglass Model to bridge high-level values and granular organisational practices, ensuring that principles such as fairness and accountability are embedded in everyday decision-making. Similarly, ref. [62] propose a unified framework that connects governance dimensions across socio-technical contexts, while [63] layered model underscores the need for oversight spanning technical design, institutional processes, and societal impacts to address information asymmetries among developers, policymakers, and citizens. Across these approaches, a consistent insight emerges; effective AI governance requires actionable structures that integrate principles, procedures, and context.

Sector-specific research further reinforces this need. Studies in healthcare governance [64] and local government technology integration [65] demonstrate that governance frameworks must extend beyond technical controls to incorporate public-interest values. At the municipal level, governance mechanisms must be iterative, participatory, and adaptive, rather than static compliance tools, particularly given the evolving ethical, social, and operational risks associated with AI-enabled public services [66].

While existing local government AI governance frameworks provide useful guidance on core principles and governance dimensions, they vary considerably in how comprehensively these elements are operationalised across the AI lifecycle [63]. Many toolkits articulate high-level commitments to transparency, accountability, stakeholder engagement, and ethical oversight, yet offer limited guidance on when, by whom, and through which procedural decision points these considerations should be enacted during AI design, procurement, and deployment [62]. In practice, this often results in governance mechanisms that are applied retrospectively, fragmented across departments, or reduced to compliance checklists rather than embedded within routine administrative decision-making [67].

For local governments, the absence of clear procedural sequencing and decision thresholds complicates the translation of normative commitments into consistent and auditable actions across service domains [65]. This gap between principled guidance and procedural execution underscores the need for a governance mechanism that integrates ethical and risk considerations into structured decision-making throughout the AI lifecycle.

5. Results

5.1. A Proposed Stage-Gate Framework for Responsible Municipal AI Governance

A stage-gate model offers a practical foundation for addressing this challenge. Originally developed by [68] for innovation and project management, the stage-gate approach divides complex initiatives into successive stages, including concept development, feasibility assessment, system development, testing, and deployment, separated by formal review gates at which projects may proceed, pause, or terminate. In private-sector contexts, the model has been shown to enhance risk management, transparency, and resource allocation by providing structured opportunities to evaluate technical, commercial, and regulatory readiness [69].

The stage-gate approach has since been adapted for public-sector initiatives, including digital transformation projects, due to its capacity to strengthen governance, accountability, and stakeholder alignment [70]. Its sequential structure enables systematic quality assurance, oversight of public resources, and disciplined management of uncertainty, all of which are critical when deploying emerging technologies such as AI.

For local governments, these principles are particularly salient. Municipalities frequently operate under conditions of constrained capacity, political sensitivity, and heightened expectations for transparency and public trust [24]. A stage-gate framework responds to these conditions by decomposing the AI lifecycle into manageable phases, embedding governance, risk assessment, and stakeholder engagement at each formal decision point. This structure supports early risk identification, aligns AI initiatives with community values, and enables adaptive responses to emerging ethical or regulatory concerns. By making responsibilities for approval and escalation explicit, the model also creates an auditable decision trail suitable for public accountability.

Within this framework, different governance concerns are addressed at stages where they are most consequential. Data governance issues, including data quality, consent, and stewardship, can be assessed at early design gates prior to system development. Procurement-related risks, such as long-term vendor dependency or contractual lock-in, are more appropriately examined at mid-stage gates when sourcing and contractual decisions are formalised. Issues related to algorithmic bias, transparency, and social impact can be evaluated at later-stage gates through testing, technical verification, and pre-deployment review. In this way, ethical and governance considerations are operationalised across the lifecycle without requiring a separate risk taxonomy.

While the stage-gate concept is adapted from established innovation and governance literature, the configuration of governance stages, decision gates, and the systematic mapping of ethical, institutional, and procedural considerations across the AI lifecycle represent the authors’ integrative synthesis. This framework is not a direct replication of existing models; rather, it reflects a conceptual contribution that combines insights from literature, policy analysis, and documented municipal practices to develop a context-sensitive governance structure tailored to local government realities.

Overall, a stage-gate AI governance framework combines ethical grounding with operational pragmatism. It enables local governments to translate abstract principles into context-sensitive, procedural practices, supporting responsible innovation while safeguarding public values and community trust.

5.2. Operational Benefits and Applicability of Stage-Gate Governance in Municipal AI

The proposed stage-gate framework offers several advantages that make it particularly suitable for governing AI adoption in local government contexts. First, its structured, sequential approach establishes clear milestones and decision points, helping municipalities navigate the complexity of AI projects with discipline, transparency, and accountability [71]. Each gate functions as a formal checkpoint to assess progress, identify risks, and evaluate ethical, legal, and social implications. This iterative review process is designed to minimise harms while maximising benefits across the AI lifecycle.

Second, the framework’s modular and adaptable design supports implementation in diverse municipal contexts [72]. Local governments differ widely in resources, institutional capacity, and community priorities. The stage-gate model allows customisation of governance activities at each stage, ensuring both small and large municipalities can adopt measures appropriate to their operational context and local needs [73].

Third, the framework promotes cross-functional coordination by requiring input from multiple departments, including IT, policy, legal, and operational units [74]. This integrated approach supports more informed decision-making and helps ensure that technical development is aligned with administrative, legal, and social considerations [37].

Fourth, by organising AI adoption into iterative stages, the framework supports adaptive management in response to evolving technologies and uncertainties [74]. Municipalities can revise, pause, or redirect initiatives based on emerging evidence, regulatory changes, or stakeholder feedback.

Fifth, evidence from public sector applications demonstrates that gate-based governance can improve resource allocation, increase project success rates, and support organisational learning [70]. Applied to local government AI adoption, the approach enables a balance between innovation objectives, ethical stewardship, democratic accountability, and risk management.

Nonetheless, despite these advantages, the stage-gate approach presents several limitations in public-sector settings. Highly structured governance processes may reinforce bureaucratic rigidity, particularly in resource-constrained municipalities where administrative burdens are already significant [75]. Repeated review cycles may also contribute to ‘gate fatigue’, reducing the effectiveness of decision-making over time. In addition, there is a risk that gate reviews become procedural rather than substantive, functioning as compliance exercises rather than meaningful evaluation processes. These challenges highlight that the effectiveness of stage-gate governance is not solely dependent on its design, but also on organisational culture, leadership commitment, and the extent to which review processes are empowered to influence real decision-making outcomes.

Figure 3 conceptualises a stage-gate structure for local government AI governance, highlighting key decision points and feedback loops that underpin responsible implementation.

Table 1. Checklist of stage-gate framework.

Gate, Stage, Reference	Description	Strategic Objective	Checklist
Gate 1: Enabling condition Stage 1: Organisational groundwork [37]	▪ Local governments must establish enabling conditions by forming a cross-functional. Build readiness through AI literacy, ethics alignment, and collaborative structures.	▪ Assemble multidisciplinary team ▪ Shared understanding of AI opportunities & risks ▪ Promote openness and accountability ▪ Define roles and responsibilities ▪ Build baseline literacy	▪ Team appointed ▪ Roles documented ▪ Collaboration protocols ▪ AI briefing paper ▪ Introductory training
Gate 2: Identifying the task Stage 2: Initial screening [76]	▪ Structured screening ensures AI is the right solution, avoids hype-driven adoption, and considers non-AI alternatives.	▪ Establish clear decision process ▪ Align with public value ▪ Avoid hype/pressure ▪ Consider simpler alternatives	▪ Suitability assessment ▪ Data readiness check ▪ Risk–benefit analysis ▪ Pre-screening form ▪ AI vs. non-AI guide
Gate 3: Ethical screening Stage 3: Ethical governance [44]	▪ Ethics must align with global standards and local values. Principles should be embedded into policies, procurement, and evaluation.	▪ Develop ethics charter ▪ Conduct impact assessments ▪ Public engagement ▪ Institutionalise ethics bodies ▪ Train staff on ethics	▪ Ethics charter ▪ Impact assessment tools ▪ Engagement toolkit ▪ Training module ▪ Advisory board terms of reference
Gate 4: Detailed investigation Stage 4a: Financial planning [37]	▪ AI governance requires sustainable budgeting, external funding, and shared resource models.	▪ Integrate AI into budgets ▪ Secure external support ▪ Lifecycle cost analysis ▪ Explore shared services ▪ Build business case for value	▪ Budget template ▪ Cost–benefit framework ▪ Shared procurement agreements ▪ Funding guide ▪ Grant roadmap
Stage 4b: Institutional capacity [37,44]	▪ Move beyond ad hoc teams. Establish permanent roles, structures, and partnerships with academia and civic tech.	▪ Define leadership roles ▪ Create governance structures ▪ Train staff ▪ Build partnerships ▪ Coordinate across departments	▪ Role descriptions ▪ Governance models ▪ Literacy training ▪ Partnership frameworks ▪ Governance handbook
Stage 4c: Data governance [44,76]	▪ AI depends on robust data governance and updated infrastructure ensuring privacy, security, and interoperability.	▪ Audit systems ▪ Update policies ▪ Clarify stewardship ▪ Upgrade infrastructure ▪ Promote open data	▪ Data governance charter ▪ Consent/sharing protocols ▪ IT audit reports ▪ Ethical data guidelines ▪ Open data platforms
Gate 5: Decision-making Stage 5: Delivery model [77]	▪ Delivery options (in-house, outsourced, collaborative) have implications for cost, oversight, and sovereignty.	▪ Match model to capacity ▪ Ensure transparency ▪ Protect data ▪ Retain control ▪ Encourage innovation partnerships	▪ Comparative framework ▪ Procurement templates ▪ Capability audit ▪ Oversight checklist
Gate 6: AI system design Stage 6: Innovation/testing [44,76,78]	▪ Safe experimentation through sandboxes, hackathons, and co-design enables learning, feedback, and adaptation.	▪ Establish sandboxes ▪ Co-create with stakeholders ▪ Support civic innovation ▪ Embed reflection cycles ▪ Share knowledge	▪ Sandbox protocols ▪ Innovation labs ▪ Grants for testing ▪ Feedback channels ▪ Co-design toolkits
Gate 7: Risk assessment Stage 7: Mitigating risks [79]	▪ Local governments must address bias, privacy, transparency, and harm risks. High-risk systems may need restrictions or bans.	▪ Classify risks ▪ Mandate algorithmic impact assessments ▪ Safeguard privacy ▪ Citizen complaints mechanisms ▪ Restrict high-risk uses	▪ Risk matrix ▪ Algorithmic impact assessment toolkit ▪ Redress protocols ▪ Ethics/safety reviews ▪ Risk decision trees
Gate 8: Policy assessment Stage 8: Embedding governance [37]	▪ AI governance must be legally grounded, aligning with national/global standards, and codified in procurement and data policies.	▪ Update frameworks ▪ Align policies ▪ Mandate transparency ▪ Ethical procurement ▪ Local codes reflecting values	▪ Governance policy/code ▪ AI clauses in contracts ▪ Data law amendments ▪ Accountability frameworks ▪ Disclosure policies
Gate 9: Implementation Stage 9: Monitoring [76,78]	▪ Implementation is iterative. Requires pilots, monitoring tools, feedback loops, audits, and clear sunsetting rules.	▪ Pilot before rollout ▪ Monitor ethics/performance ▪ Gather citizen feedback ▪ Regular audits ▪ Define decommissioning criteria	▪ Deployment roadmap ▪ Monitoring dashboard ▪ Feedback channels ▪ Audit reports ▪ Sunset protocols
Gate 10: Evaluation Stage 10: Evaluation [76]	▪ Evaluation checks whether AI meets objectives and ethics, informing scaling, redesign, or decommissioning.	▪ Post-implementation evaluation ▪ Identify successes/challenges ▪ Guide redesign/scaling ▪ Transparent reporting ▪ Feed results into early stages	▪ Evaluation framework ▪ Success criteria checklist ▪ Data templates ▪ Report template ▪ Loopback mechanisms

complements this framework by translating each stage and gate into a practical checklist informed by prior studies. The checklist guides municipalities in assessing organisational readiness, embedding ethical considerations, and engaging stakeholders throughout the AI lifecycle.

For example, the first stage (organisational groundwork) may include: (a) Appointing a multidisciplinary team; (b) Defining roles, responsibilities, and decision-making authority; (c) Establishing cross-department communication protocols; (d) Preparing an internal briefing on AI applications, risks, and relevance for municipal operations; (e) Conducting introductory AI literacy workshops. At Gate 1, these activities are evaluated to determine whether the project may proceed (‘go’), requires refinement (‘hold’), or needs substantial revision. This ensures that enabling conditions and organisational readiness are established before AI initiatives advance.

In practice, municipal experiences illustrate how the framework’s stages map onto real-world governance (

Table 2. Municipal case studies illustrating governance gate practices.

Gate	Example Local Government	Practice	Insight
Gate 1	City of Greater Dandenong, Australia [57]	Discovery phase: stakeholder workshops, system mapping, readiness assessment	Cross functional collaboration and organisational readiness building for AI adoption
Gate 2	Maidstone Borough Council, UK [80]	Predictive analytics (OneView) for homelessness risk (early intervention model)	Clear problem definition and proactive, prevention focused AI use case aligned with public value
Gate 3	Sydney, Australia (City of Sydney/NSW context) [81]	Ethical governance and AI in planning resolution	Transparent policies for AI decision making ensured fairness, accountability, and trust among stakeholders
Gate 4	Finney County, USA (Finney, Seward, and Ford Counties) [82]	Regional collaboration and leadership buy in for AI enabled foot traffic analytics	Financial planning and institutional capacity via shared cost models, regional collaboration, and leadership support
Gate 5	City of Pretoria, South Africa [83]	AI driven digital twin using existing geospatial and waste management data	Delivery model choice that leverages existing infrastructure and internal capacity while retaining control over data and implementation
Gate 6	Sydney, Australia (NSW ePlanning) [81]	Use of proven AI tools in planning workflows	Use of reliable, market-tested AI solutions accelerated implementation and ensured performance in a real-world planning environment
Gate 7	San Francisco, USA [84]	Ban on city-department use of facial-recognition technology (Acquisition of Surveillance Technology Ordinance)	Precautionary risk-mitigation measure, restricting high-risk AI-enabled surveillance where error-prone and civil-liberties-invasive systems pose disproportionate harm to rights and trust
Gate 8	Local governments with AI specific policies (e.g., City of Spokane, USA; Greater Sudbury, Canada; Shire of Northam, Australia)	Development of local AI governance policies and codes	Policy level embedding of AI governance, aligning procurement, data use, and oversight with local values and legal frameworks
Gate 9	Peterborough City Council, UK [85]	Hey Geraldine: AI-powered personalised assistant supporting social-care staff with knowledge access and feedback-driven refinement	Iterative implementation combining pilot-driven deployment, user-interaction-based analytics, and continuous improvement, reflecting feedback-loop-oriented monitoring
Gate 10	City of Tshwane, South Africa [86]	AI-driven digital twin with interactive dashboard for waste-collection performance	Post-implementation evaluation and real-time optimization, using dashboard-based performance-insights to allocate resources, adjust schedules, and improve operational-responsiveness

). The City of Greater Dandenong’s discovery-phase readiness work exemplifies organisational groundwork (Gate 1), while Maidstone Borough Council’s predictive-analytics-based early-intervention model for homelessness risk reflects a clear, prevention-oriented AI use-case (Gate 2). Sydney’s ethics-grounded AI-in-planning approach shows how transparent governance supports public trust (Gate 3), and São Paulo’s Smart-Sampa facial-recognition platform demonstrates high-threshold risk-mitigation with human oversight (Gate 7). Bristol City Council’s Briz digital assistant illustrates iterative implementation and feedback-driven monitoring (Gate 9), and São Paulo’s real-time public reporting of AI-system outcomes exemplifies post-implementation evaluation and transparency (Gate 10). These cases show how municipal practice informed the refinement of the framework’s governance stages.

5.3. Minimum Viable Stage-Gate Pathway for Resource-Constrained Municipalities

Building on the preceding discussion, a minimum viable version of the stage-gate framework can enhance its practical applicability for resource-constrained municipalities. Rather than operationalising all ten gates in full, smaller councils can prioritise a core subset of critical governance checkpoints: (a) Enabling conditions (Gate 1); (b) Initial screening (Gate 2); (c) Ethical screening (Gate 3); (d) Delivery model decision-making (Gate 5), and; (e) Risk assessment (Gate 7). Together, these gates address foundational requirements for readiness, legitimacy, and risk mitigation. Subsequent stages (such as monitoring, evaluation, and continuous improvement) can be implemented in simplified formats or through shared service arrangements. This tiered configuration enables incremental adoption, allowing municipalities to balance governance rigour with institutional capacity constraints.

With appropriate adaptation to local governance cultures, resource endowments, and stakeholder expectations, the stage-gate framework offers a structured and democratically legitimate pathway for municipal AI adoption. Its implementation does require upfront investment in governance systems, capability development, and organisational change; nevertheless, such investments are increasingly indispensable given the expanding scope and societal implications of AI. The Victorian Government’s Rural Councils Transformation Program provides an indicative example of how gate-based digital governance approaches can be tailored to local government contexts [87].

In sum, the stage-gate framework integrates structured sequencing, contextual flexibility, and participatory oversight to support responsible municipal AI deployment. It translates complex innovation processes into manageable, transparent, and ethically grounded steps, enabling municipalities to pursue innovation while maintaining public trust.

6. Discussion

6.1. Strategic Pathways for Implementing Responsible AI Governance

Operationalising responsible AI governance in local government requires a coordinated set of context-sensitive strategies. Drawing on policy literature, empirical case studies, and observed implementation challenges across jurisdictions, six interrelated recommendations emerge [88]. Collectively, these strategies provide a structured yet adaptable approach to build institutional capacity, embed ethical safeguards, and strengthen public trust in municipal AI initiatives. Responsible governance also entails recognising that AI adoption is contingent rather than inevitable; frameworks should therefore support both informed restraint and implementation [89]. Figure 4 synthesises these recommendations within a broader local AI governance framework.

The first step is a systematic assessment of local government readiness. Prior to developing or deploying AI tools, municipalities should use structured diagnostic approaches to evaluate capabilities, identify governance gaps, and assess risks and ethical alignment [90]. These assessments must also consider social priorities, workforce implications, and environmental constraints, rather than focusing solely on technical feasibility [91]. A graduated AI maturity model provides a useful baseline, defining progressive capability levels across governance structures, technical infrastructure, ethical oversight, and public engagement, enabling benchmarking, incremental goal setting, and avoidance of premature deployment of complex AI systems [77]. Recent work, such as [92], further reinforces the relevance of structured readiness frameworks in local government AI governance contexts. This corresponds primarily to early-stage governance functions within the stage-gate framework, particularly enabling conditions and initial screening (Gates 1–2), which establish organisational readiness and decision clarity prior to AI adoption.

Embedding ethical governance mechanisms early in the AI lifecycle is central to responsible adoption. Ethical criteria, transparency obligations, and accountability provisions should be integrated into procurement processes, vendor contracts, and technical design specifications [93]. Procurement and contracting shape power relations between local governments and technology providers, influencing data ownership, algorithmic control, and vendor dependency [1]. Evidence indicates that ethics by design reduces costs associated with retrofitting AI systems, ensures alignment with societal values, and strengthens public legitimacy [19]. Within the stage-gate framework, these functions align with ethical screening and governance integration stages (Gate 3), as well as later policy and risk management checkpoints (Gates 7–8).

Active public engagement is normatively and instrumentally critical. Participatory co-design workshops, citizens’ juries, and deliberative forums help ensure socially acceptable outcomes and enhance legitimacy [94]. Inclusiveness prevents disproportionate impacts on marginalised communities and ensures AI initiatives reflect diverse community perspectives [95]. Engagement also surfaces labour, service quality, and public-sector workforce concerns that may be reshaped or displaced by AI automation [96]. This cuts across multiple stages of the framework, particularly ethical governance (Gate 3), co-design and testing (Gate 6), and monitoring and evaluation stages (Gates 9–10).

AI governance requires continuous monitoring and accountability. Oversight mechanisms, including audits, dashboards, and algorithmic impact assessments, enable municipalities to detect unintended harms, respond to social and environmental changes, and maintain transparency through iterative review [97]. Accountability is best conceived as an ongoing governance cycle embedded in institutional practice, rather than a one-off compliance milestone [98]. This approach is particularly important given AI’s material and environmental footprint, including energy-intensive computation, data centre reliance, and downstream impacts such as e-waste [99]. These functions directly correspond to the later stages of the stage-gate framework, especially monitoring and evaluation (Gates 9–10), where ongoing oversight and iterative learning are institutionalised.

Inter-local collaboration emerges as a transversal enabler of responsible governance. Peer networks, shared procurement platforms, and intergovernmental partnerships facilitate knowledge sharing, resource efficiency, and adoption of best practices [100]. Smaller or resource-constrained municipalities benefit disproportionately from such arrangements. Collaboration aligns with commons-based and cosmolocalism governance models, leveraging open-source tools, shared standards, and collectively developed practices to reduce reliance on proprietary platforms [101]. Municipal data sovereignty initiatives, such as Barcelona’s, demonstrate how local governments can treat public data and algorithms as civic assets rather than vendor-controlled resources [102]. Within the framework, collaboration supports delivery model decisions and institutional capacity development (Gates 4–5), enabling shared governance and resource optimisation.

Together, the six strategic actions mentioned in Figure 4 (i.e., continuous accountability, readiness assessment, ethical design, procurement and vendor governance, public engagement, and intergovernmental collaboration) provide a pragmatic toolkit for governing AI responsibly within institutional and resource constraints. These actions are not conceived as standalone measures; rather, they map directly onto stages and gates within the proposed framework, reinforcing its operational coherence and practical applicability. The framework is compatible with conventional governance arrangements as well as alternative models prioritising open-source, commons-based, and sustainability-oriented approaches. It explicitly recognises the contested political, ecological, and labour terrain in which local government AI decisions are made.

By operationalising these actions, municipalities can exert agency over AI development and deployment, safeguard citizens, foster innovation, and navigate complex sociotechnical challenges without treating technology adoption as an end in itself. While conceptually grounded and operationally structured, the framework requires empirical testing through real-world case studies to evaluate its effectiveness, limitations, and adaptability across diverse municipal contexts.

6.2. Limitations and Research Directions

This study develops a conceptual framework and stage-gate model for responsible municipal AI governance; however, several limitations should be acknowledged. The analysis is primarily qualitative and conceptual, drawing on selected illustrative cases and global-to-municipal governance literature. As such, the framework has not been empirically tested in real-world municipal settings, which limits its demonstrated applicability and generalisability across diverse local government contexts. In addition, its effectiveness is likely to vary depending on differences in institutional capacity, regulatory environments, and political and administrative culture, particularly in smaller or resource-constrained municipalities. The study also reflects a temporally bounded synthesis of a rapidly evolving technological and regulatory landscape and does not fully capture the dynamic interactions between local, regional, and national governance systems over time.

These limitations suggest several directions for future research. First, empirical validation through comparative case studies, pilot implementations, or action research in local government settings is needed to assess the practical feasibility and effectiveness of the proposed framework. Second, longitudinal studies could examine how municipal AI governance structures evolve in response to technological change, regulatory development, and institutional learning over time. Third, participatory and mixed-methods research could provide deeper insight into how different stakeholders, including citizens, public servants, and technology providers, engage with and shape AI governance processes in practice. Finally, future work could translate the framework into operational tools such as governance maturity models, audit instruments, or evaluation metrics to support evidence-based implementation and cross-jurisdictional comparison.

7. Conclusions

This paper argues that a stage-gate approach offers a practical way for local governments to translate high-level AI principles into structured, reviewable governance decisions across the AI lifecycle. In doing so, it responds to the need for a context-sensitive model of municipal AI governance that aligns innovation with accountability, public value, and institutional capacity.

The study makes two key contributions.

First, it offers a conceptual contribution by emphasising why municipal AI governance requires a framework tailored to the institutional realities of local government rather than adapted uncritically from national or corporate contexts. It provides a pragmatic structure that integrates ethical oversight, risk management, and community engagement while recognising local capacity constraints and accountability obligations.

Second, it offers an operational contribution, framing the model as both protective and enabling. Protectively, it introduces decision checkpoints that reduce reliance on vendors, mitigate risks from opaque algorithms, and prevent ethical and operational failures that could erode public trust. Enabling this, it embeds democratic values, inclusiveness, and ethical safeguards into AI adoption processes, positioning local governments as active stewards rather than passive implementers. By structuring the AI lifecycle into distinct stages, the model creates repeated opportunities for ethical reflection, stakeholder engagement, and technical oversight, equipping municipalities to align innovation with accountability.

In addition, the paper identifies a set of priority questions for future empirical work to examine the framework’s feasibility, scalability, and adaptability across diverse municipal contexts. These questions include:

▪

How can high-level AI ethics principles be operationalised into practical tools (such as risk assessments, procurement guidelines, and audit protocols) that are realistic for resource-constrained local governments?

▪

Which participatory approaches best ensure sustained, representative community involvement, particularly in contexts where public input is fragmented or dominated by powerful actors?

▪

How can procurement practices and vendor relationships be structured to prevent lock-in while safeguarding long-term data stewardship, transparency, and accountability?

▪

What forms of inter-local collaboration (regional consortia, peer networks, or cross-border partnerships) most effectively support smaller municipalities in building capacity, legitimacy, and knowledge?

▪

Which metrics best evaluate AI governance outcomes beyond technical efficiency, capturing both legitimacy and ethical effectiveness?

▪

How can city-specific conditions, such as density, socio-spatial inequality, infrastructure complexity, and environmental vulnerability, be integrated into municipal AI governance to ensure context-sensitive and equitable outcomes?

Answering these questions will refine the stage-gate framework and assess its utility as both an analytical and practical instrument for municipal governance. Comparative research across jurisdictions will be particularly valuable in understanding how social, cultural, and institutional factors shape implementation outcomes.

A structured governance framework extends beyond risk mitigation. It enables municipalities to balance innovation with ethics, efficiency with fairness, and technological opportunity with public accountability. By proactively engaging with AI governance, rather than reacting to technological pressures, local governments can contribute to trusted and democratic smart city agendas while reinforcing their agency in shaping inclusive, sustainable, and legitimate urban futures.

The framework is designed for broad applicability across cities, yet its implementation must remain sensitive to local urban contexts. Large or well-resourced cities may operationalise the model through dedicated urban AI units, formal cross-departmental review committees, and advanced auditing and monitoring mechanisms. In contrast, smaller cities or resource-constrained municipalities may rely on streamlined or collaborative arrangements, such as shared metropolitan review bodies, regional procurement platforms, or the selective application of critical governance gates for high-impact urban use cases. In such contexts, a minimum viable configuration of the framework may prioritise a subset of core gates, such as enabling conditions (Gate 1), initial screening (Gate 2), ethical screening (Gate 3), delivery model decision-making (Gate 5), and risk assessment (Gate 7), while later-stage functions (e.g., monitoring and evaluation) may be implemented in shared institutional arrangements (i.e., regional collaboration or external support). Accordingly, the stage-gate model is adaptive rather than prescriptive, enabling cities to tailor governance processes to their institutional capacity, socio-spatial conditions, and service priorities while maintaining structured oversight, ethical integrity, and public accountability in urban AI deployment.