This chapter examines Estonia’s aim to transition to a proactive state driven by AI, contrasting its high-level policy ambitions with ground-level operational realities. Estonia has built a solid foundation of digital government. However, the OECD innovation capacity data exposes an implementation gap: despite Estonia’s reputation as a digital pioneer, its decentralised, bottom-up approach to generative AI has contributed to fragmented governance, uneven data quality, and duplicated efforts. As in many countries, while frontline innovators are eager to adopt AI, they are throttled by rigid public procurement, a risk-averse legal culture, and middle management capability deficits. By identifying these systemic bottlenecks, the OECD recommends balancing technological aims with a more comprehensive approach to digital literacy, process streamlining, and strategic training.
Strengthening the Innovative Capacity of the Public Sector of Estonia
7. The emerging field of AI
Copy link to 7. The emerging field of AIAbstract
Strategic aims and objectives
Copy link to Strategic aims and objectivesEstonia aims to create a proactive state driven by AI, built on a solid foundation of digital government. The 2025 edition of the OECD Digital Government Index scored Estonia 0.83 overall, above the OECD average of 0.70, and high on Data-driven Public Sector (0.93) (OECD, 2026[1]). This strong performance suggests that Estonia’s AI ambitions can be built on mature digital government foundations.
Estonia aims to transition from collecting data to operating a proactive, predictive state. This vision relies on data-based decision-making to boost government and business productivity. decentralised infrastructure, growing domestic capacity and timelines articulated in various action plans. The Digital Society Development Plan 2035 governs the digital state over the next decade (Ministry of Justice and Digital Affairs, 2026[2]). The White Paper on Data and Artificial Intelligence 2024-2030 focuses specifically on data assets and algorithmic principles (Government of Estonia, 2024[3]). Estonia’s Digital Agenda 2030 articulates the government’s vision for how digital technology, including Generative AI, will develop the Estonian economy, state and society (Ministry of Economic Affairs and Communications, 2021[4]). These frameworks use the European Union AI Act as a regulatory baseline. They implement the Act’s risk-based compliance hierarchy across public sector AI applications.
Beneath these long-term frameworks sit operational toolkits like the Data Action Plan 2024-2025 (Ministry of Economics and Communication, 2024[5]), White Paper on Data and Artificial Intelligence 2024-2030 (Government of Estonia, 2024[3]), and the Artificial Intelligence Action Plan 2024–2026 (Government of Estonia, 2024[6]). This structural division ensures that while the state's societal goals and objectives for trustworthy AI remain steady, technical milestones can adjust quickly to rapid breakthroughs in large language models. To prevent fragmentation across ministries, the Digital Society Development Plan 2035 introduces centralised pillars to coordinate public AI efforts:
National AI applications catalogue: a shared repository allowing state agencies to reuse algorithms, preventing redundant procurement and development costs.
Domestic computing capacity: a targeted initiative to build out domestic public sector computing power and native Estonian language models, explicitly minimising dependency on foreign commercial cloud providers.
Unified digital gateways: Upgrading the central portal (eesti.ee), where citizens interact with Government (e.g. accessing registrations, prescriptions, applying for services, etc.), and the national mobile application to serve as the main secure access points for AI-driven notifications and verified digital identity systems, including the integration of the European Digital Identity (EUDI) wallet.
In the Digital Society Development Plan 2035 (Ministry of Justice and Digital Affairs, 2026[2]), Estonia is positioning itself to enter the era of agentic AI. The primary paradigm shift moves public services away from old query-based interfaces. Instead, a user can request a complex service such as ordering a new passport, and an AI agent orchestrates the interaction across the relevant authority. This moves complexity into the background through system-to-system interaction while allowing the user to validate key actions where needed. (OECD, 2026[1]) . Furthermore, the Development Plan and the White Paper on Data and Artificial Intelligence 2024-2030 (Government of Estonia, 2024[3]) share a unified vision for life-event and business-event service chains. This aligns with the Digital Government Outlook’ finding that mature digital government systems increasingly need to move from digitising services to redesigning service delivery around user needs, data reuse and proactive interaction (OECD, 2026[1]).
Estonia’s current public sector AI governance setup combines horizontal and vertical coordination mechanisms with more advanced mainstreaming arrangements across government. Horizontal coordination within the Ministry of Justice and Digital Affairs supports alignment on cross-cutting digital and AI-related priorities, while vertical coordination links central policy direction with implementation at the level of individual ministries and their domains. Digital development is embedded into sectoral policy and service development as well as treated as a separate policy area, allowing responsibilities to remain distributed while common priorities, standards and enabling capabilities are coordinated centrally. The support is also available to facilitate frontrunners to support community learning as well as building and sharing best practices.
While a governance system and ambitious goals exist, the interviewed experts argued that currently the implementation structure itself remains inadequate. The Ministry of Justice and Digital Affairs is reported to manage numerous initiatives with minimal staff. Interviewees reported that staff shortage severely limited their ability to drive changes across a highly decentralised digital government. The Government Office recently elevated AI to the central government level (see Box 7.1). The Prime Minister actively supports these initiatives, and initial pilots are underway (Government Office, 2026[7]). However, given the current state of innovation capacity and AI literacy (see discussion below), these resources will likely be insufficient.
Box 7.1. Estonia.ai programme
Copy link to Box 7.1. Estonia.ai programmePrime Minister Kristen Michal launched the Eesti.ai (Estonia.ai) programme in January 2026. The Government Office created this initiative in partnership with leading entrepreneurs. The programme currently finalises its exact funding requirements for the spring and summer of 2026. It targets significant operational milestones over an initial 18-month period. Long-term economic goals align with a 2035 target.
Eesti.ai aims to systematically implement artificial intelligence across all sectors. The primary objective is to double the value of Estonian labour by 2035. The government expects this systematic adoption to drive significant economic expansion. Targets include growing the economy by 25% within five years and 50% within ten years. This translates to an estimated €20 billion in economic growth. Ultimately, Estonia aims to become a global leader in AI implementation.
The Government Office manages the Eesti.ai initiative. An international advisory council reports directly to the Prime Minister. The board includes technology entrepreneurs, international experts, and start-up sector representatives. A cross-ministerial working group also supports the programme's governance.
The initiative mainly focuses its resources on practical, high-impact implementation. Currently the council has approved 15 high-impact public-private projects. These initiatives focus on healthcare, education, security, and industry. The programme prioritises practical skills. It plans to train at least 100 000 members of the public through free AI workshops over 18 months. As such the upskilling initiatives target the broader working-age population and are not designed specifically for public sector employees. Eesti.ai also hopes to scale existing successful initiatives. This includes expanding the educational AI Leap programme and public sector AI applications. Lastly, the initiative collects public suggestions. Citizens submit everyday problems where AI could streamline processes or save time.
Effective AI integration requires cross-cutting business lines that break existing bureaucratic silos. However, interviewed experts express concern about the financial viability of this approach. For example, existing life-event services lack assured long-term sustainability due to missing funding models. The state often relies on external project grants that focus solely on the development stage (see Chapter 2). This approach neglects the future management and maintenance costs of digital services. Consequently, authorities struggle to find state budget funding to keep solutions functional once initial project grants end. This might also be a considerable issue in the case of AI.
Estonia's reliance on external project grants and its ambition to build domestic AI computing capacity both raise a further question beyond financing: whether the public sector retains sufficient in-house capability to act as an informed buyer and steward of externally developed or co-financed AI solutions. Forthcoming OECD evidence identifies this build/buy/borrow decision as consequential for the long-term depth and resilience of public sector AI capability, regardless of the sourcing model chosen (OECD, forthcoming[10]).
Currently, economic commitments outlined in recent strategies mark a major scale-up in state investment. The Digital Society Development Plan 2035 (Ministry of Justice and Digital Affairs, 2026[2]) forecasts that Estonia will require approximately EUR 1.8 billion in total funding between 2026 and 2035 to build out this automated governance infrastructure within the Ministry of Justice and Digital Affairs. At the same time, the widespread, structured integration of AI systems across public bodies is projected to lower state service, development, and administrative workflow costs by at least EUR 425 million over that same period. Furthermore, AI deployment is viewed as a vital macroeconomic tool to offset demographic decline and help sustain annual economic growth above 5% (Ministry of Justice and Digital Affairs, 2026[2]), challenging many ministries and policy areas to come up with relevant innovations.
Financing (outside of distinct pilots) is closely tied to public upskilling and regulatory measures:
The "AI Leap 2025" initiative (Government of Estonia, 2024[6]): this programme embeds advanced algorithmic concepts and digital pedagogy directly into secondary and technical school curricula to ensure future graduates are prepared for an AI-driven workforce.
Public sector data literacy: the White Paper on Data and Artificial Intelligence 2024-2030 (Government of Estonia, 2024[3]) outlines ongoing training for public managers. These programmes are designed to give civil servants the skills to manage algorithmic procurement, spot automation opportunities, and run mandatory data-privacy impact assessments.
Co-financing and regulatory sandboxes: As detailed in the Data Action Plan 2024-2025 (Ministry of Economics and Communication, 2024[5]), the state aims to use targeted financial grants and secure sandboxes to encourage domestic technology firms to build public sector solutions, ensuring that state investments directly stimulate the local tech ecosystem. Partnerships with Finland for a joint approach on this are on the way (Finnish Government, 2026[11]).
The Estonian government identifies significant technical, operational, international, and legal challenges affecting implementation. The Data Action Plan 2024-2025 notes highly uneven data quality across the public sector (Ministry of Economics and Communication, 2024[5]). Many local government registries use legacy software failing to meet national standardisation or interoperability requirements. Upgrading historical datasets and ensuring continuous validation requires major funding. This diverts resources away from innovation with AI. Operating a hyper-digitised state introduces severe vulnerabilities amid rising challenges from state and non-state actors (Ministry of Justice and Digital Affairs, 2026[2]). Essential services like healthcare and identity verification are increasingly automated and data-dependent. They present high-value targets for hostile cyber actions. The Artificial Intelligence Action Plan 2024–2026 notes the heavy administrative burden of the EU AI Act (Government of Estonia, 2024[6]). Strict documentation, auditing, and oversight requirements can make state agencies more risk-averse. The long-term strategy prioritises ensuring citizens understand automated decisions and maintain contestation rights.
A digital backlash remains possible if algorithmic decisions lack human oversight or transparency. The OECD Survey on Drivers of Trust in Public Institutions found that 35% of Estonians believe the government will use AI to provide more tailored services. One-third think AI will be used to reduce costs. Almost one-in-four are confident the government will protect personal information when using AI; and believe the government will ensure fair treatment when AI is deployed in public services. Estonians were less positive about the potential use of AI by government agencies than the OECD average (OECD, 2026[12]).
The recent OECD Governing with Artificial Intelligence (2025[13]) report, outlines similar challenges across OECD countries and beyond. AI has the potential to boost productivity and transform public service delivery. However, public administrations face data governance deficits, digital skill shortages, and absent institutional guardrails. These gaps increase risks like skewed data in AI systems that can lead to adverse outcomes for some individuals or groups, or accountability failures. Current public sector AI adoption features fragmented and isolated piloting. Departments frequently launch small-scale trials without shared infrastructure or a unified strategic vision. Governments struggle to scale successful projects. This leads to duplicated efforts, inefficient resource allocation, and non-interoperable systems. These issues ultimately hinder widespread digital transformation. To counter this, the OECD Framework for Trustworthy AI in Government (Figure 7.1) outlines the importance of addressing different enablers, engagements and guardrails in the public sector. The innovation capacity assessment particularly looks at the skills and talent components and broader government processes in integrating AI into the public sector. The assessment does not evaluate the engagement and guardrails components of the framework; nor does it assess broader investments, infrastructure and data policies connected to the adoption of AI.
Figure 7.1. OECD Framework for Trustworthy Artificial Intelligence in Government
Copy link to Figure 7.1. OECD Framework for Trustworthy Artificial Intelligence in GovernmentThe state of generative AI use
Copy link to The state of generative AI useThe innovation capacity assessment includes a module on generative AI (GenAI), a sub-set of AI technologies that has demanded significant global attention since late 2022. Survey results indicate local and central government innovators use GenAI with similar frequency (Figure 7.2) – this question asked those public servants who had innovated in past 24 months “Thinking about this innovation, did it use or not use Generate Artificial Intelligence (GenAI)?”. GenAI adoption varied significantly across public sector organisations. AI literacy skills heavily drive generative AI adoption across all government levels (Figure 7.3) – this reflects the self-rated confidence of public servants in applying AI for innovation, see Chapter 5. Most organisations lack sufficient governance frameworks for GenAI. They will need to expand both their usage capacity and governance maturity (Figure 7.4) – where governance maturity reflects the share of 'Yes' responses to whether policy, guideline, and GenAI support questions among surveyed heads of organisation.
Figure 7.2. Generative AI use across government levels
Copy link to Figure 7.2. Generative AI use across government levelsProportion of innovators who used generative AI tools in their innovation projects, comparing central and local government.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
Figure 7.3. Generative AI adoption maturity matrix
Copy link to Figure 7.3. Generative AI adoption maturity matrixEach dot represents an organisation, coloured by government level. Dashed lines show overall means.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
Figure 7.4. AI governance maturity matrix
Copy link to Figure 7.4. AI governance maturity matrix
Note: 2x2 matrix plotting organisations by their AI governance maturity (Y-axis) against usage capacity (X-axis). Governance maturity is calculated as the proportion of 'Yes' responses from heads of organisations to policy, guideline, and GenAI support questions. Usage capacity combines GenAI adoption and staff AI skills confidence scores. Only organisations with leadership survey responses are included.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
When it comes to AI literacy of individual public servants, staff under 40 years old tend to show higher AI literacy, while staff over 60 years old show substantially lower AI literacy (Figure 7.5). Managers have higher confidence in their skills-sets using AI tools across the board, however, even them are not generally confident in their ability to build institutional pathways for public AI engagement (Figure 7.6). These patterns echo a broader finding from the forthcoming OECD paper on bridging AI skills and talent gaps in the public sector (OECD, forthcoming[10]): skills needs differ systematically by role, and leadership capability – both the AI literacy leaders need for themselves and their capacity to enable others – is consistently underexplored relative to frontline technical skills.
Figure 7.5. Reported AI literacy – staff (non-managers)
Copy link to Figure 7.5. Reported AI literacy – staff (non-managers)Deviation from neutral (3) on 5-point scale, by age group.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
Figure 7.6. Reported AI literacy – managers
Copy link to Figure 7.6. Reported AI literacy – managersDeviation from neutral (3) on 5-point scale, by age group.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
The challenge of tying AI to government processes
Copy link to The challenge of tying AI to government processesThe innovation capacity assessment results reflect these concerns. Additionally, data maturity, lack of AI infrastructure (compute access) and Estonian language within frontier AI models are cited as barriers (Ministry of Justice and Digital Affairs, 2025[14]). AI-powered innovation remains a priority. Crucially, AI is poised to unlock its most significant opportunities not merely through technological novelty, but around non-technical and process innovation. Success with AI will ultimately be a function of fundamentally redesigning workflows, frequently across traditional institutional silos, with technology serving as an enabling layer. This necessitates a structured organisational redesign analysis – a critical strategic component that is currently neither included nor accounted for within public sector organisations. This echoes the finding of the Digital Government Outlook that data-driven government depends on the governance, standards and interoperability arrangements that allow data to be reused safely across the public sector (OECD, 2026[1]).
Compounding this challenge, adding new technology to outdated processes creates friction. Streamlining and implementation capacity need to become top priorities. Staff frequently cite streamlining processes and reducing bureaucracy as major organisational challenges next to traditional technology implementation challenges connected to data, AI etc. (Figure 7.7). However, a unique opportunity lies in the shifting nature of technology development itself. The radically lower practical barriers to 'doing tech' introduced by agentic AI could significantly accelerate the public sector's innovation rate (OECD, forthcoming[15]). For Estonia, this shift offers a potent psychological lever: by framing generative and agentic AI tools as technical building blocks, the government can leverage the traditional tech-centricity of Estonian civil servants, effectively helping to drive rapid, hands-on experimentation with workflow and process optimisation. These priorities require the wider skill sets outlined in Chapter 5.
At the same time, lowering the technical barrier to building automation carries a corresponding risk: as the forthcoming OECD evidence (OECD, forthcoming[10]) notes, over-reliance on AI-generated code and outputs can erode the underlying technical understanding needed to oversee, debug, and take responsibility for these systems. Estonia's tech-centricity civil service culture may mitigate this risk relative to other countries, but it warrants monitoring as agentic tools scale.
Figure 7.7. Frequency of organisational innovation needs
Copy link to Figure 7.7. Frequency of organisational innovation needsThe analysis is based on LLM analysis of open-ended responses to the question: “What aspect of your organisation do you think needs innovation most?”
Note: 1 172 responses from the surveys of heads of organisations and public sector employees were used in Modules 2 and 3. Responses were translated into English using the Deep Translator API and underwent human quality assurance. An OpenAI embedding model was used for semantic processing, UMAP and agglomerative clustering were used for semantic-based clustering, and Gemini Flash was used for cluster labelling.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
AI and data literacy remain highly sought after
The OECD found that 36% of survey respondents linked AI to skills they most want to improve (Figure 7.8). Strategic management and service delivery represent 28% of upskilling needs. Communication and influence account for 11%, while general digital literacy accounts for 9%. Estonia's public servants recognise the AI revolution and want to adopt it. The data shows a risk of purchasing advanced AI tools without internal project management skills. Public servants need to understand AI and know how to integrate it into daily service delivery.
Figure 7.8. Frequency of reported skills needs
Copy link to Figure 7.8. Frequency of reported skills needsThe analysis is based on LLM analysis of open-ended responses to the question: “In general, what skills would you most like to improve?”
Note: 1 172 responses from the surveys of heads of organisations and public sector employees were used in Modules 2 and 3. Responses were translated into English using the Deep Translator API and underwent human quality assurance. An OpenAI embedding model was used for semantic processing, UMAP and agglomerative clustering were used for semantic-based clustering, and Gemini Flash was used for cluster labelling.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
To realise these AI goals, greater operational and strategic support is required from the centre of government. In particular, the specialised teams within the Ministry of Justice and Digital Affairs, given its role in AI policy development and implementation co-ordination, need sufficient staff. This is also crucial within public sector organisations. Relying on isolated personnel – such as having only one individual in charge of national AI implementation – makes digital transformation fragile and creates vulnerabilities to staff turnover and knowledge loss.
Furthermore, public sector organisations need to anticipate and plan for higher operational costs during the AI implementation lifecycle. Traditional legacy systems and processes need to be run in parallel with the new AI-driven solutions to ensure continuity of public services and mitigate systemic risks. These potential duplicate running costs should be explicitly budgeted for and bridged across all public sector organisations, for example, for a minimum transition period of six months to two years. Consequently, robust Return on Investment (ROI) and cost-benefit analysis become vital tools for evaluating and justifying these lifecycle investments (see also Chapter 6 and the forthcoming report (OECD, forthcoming[16])), ensuring that the long-term efficiencies of AI offset the initial friction of parallel operations.
Overcoming data silos
Furthermore, interviewees outlined that, even though the government has clear goals to dismantle data silos in the Estonian public sector, the limitations of using registry data for policymaking were universally reported as barriers to innovation. A lack of data interoperability was also seen as a major barrier to proactive service development using the "life event" model for service delivery (a problem that municipalities also emphasised strongly). Beyond interoperability, interviewees identified the overinterpretation of data protection laws (GDPR) as a barrier. They reported that systemic fear of legal jeopardy regarding secondary data processing effectively shuts down brainstorming for proactive, cross-agency services. A pervasive culture of 'legalistic positivism' means officials will only process data if explicitly permitted by highly detailed laws, hindering the transition to an 'agentic' or proactive state. This is exemplified by the development of life and business event services, which have been in the making for almost a decade and have not yet been fully developed due to, among other factors, the strict interpretation of data protection laws.
Enabling factors for AI adoption
Copy link to Enabling factors for AI adoptionToolkits, guides, skills, and strategic alignment drive professional staff to adopt generative AI (Figure 7.9). Core resources drive managers (Figure 7.9). Frontline staff want to innovate but lack the strategic support and streamlined processes to launch ideas. The OECD (2026[17]) recently analysed Estonia’s decentralised, bottom-up approach to generative AI experimentation as part of its work on GenAI experimentation guidelines. The government avoids heavy-handed regulation (outside of complying with EU regulations) to keep pace with technological shifts. This flexible model sparks agile innovation through a culture of learning by doing. However, it creates hurdles regarding project scaling, rigid public procurement, and civil service capability gaps. The absence of a central playbook leaves departments with limited technological understanding and legal uncertainty. A decentralised model can foster a siloed, project-based mindset. This mindset increases duplicated efforts and leaves successful pilots as isolated experiments. Traditional public procurement procedures struggle to match fast-paced AI developments (Chapter 2). Inflexible funding arrangements limit the ability of teams to iterate or halt unsuccessful trials.
Furthermore, interviewed stakeholders outlined that, although public sector bodies are increasingly encouraged to apply for AI innovation funding, a critical gap exists upstream: high-stakes algorithmic applications – such as early warning systems for juvenile offending risk or tracking learner progress – frequently lack independent methodological validation (similar to the evaluation gaps for other innovations outlined in Chapter 6). When researchers are consulted, this tends to happen through informal personal networks. This collaboration lacks a joint proof of concept with a named problem owner on each side. To resolve this "interface gap", the Government Office, in partnership with academic institutions, could consider designing a structured pipeline for university expertise to enter projects at the idea formulation and evaluation stages. This framework should standardise a "joint proof of concept" format.
Figure 7.9. Key factors enabling generative AI adoption
Copy link to Figure 7.9. Key factors enabling generative AI adoptionAnalysis of which success factors enabled GenAI usage in innovation projects, by role type.
Source: OECD based on the Public Sector Innovation Capacity Survey (2026).
Recommendations
Copy link to RecommendationsEstablish an explicit, cross-agency regulatory and operational framework for data processing (Ministry of Justice and Digital Affairs, and Government Office). While Estonia has created world leading data-exchange infrastructure, there are operational, organizational, and regulatory frictions that hinder its transition to a proactive, agentic state. To affect this transition:
The Ministry of Justice and Digital Affairs, in collaboration with relevant digital and data protection authorities, should evaluate how legislation, regulation and soft-law can be updated to enable great data interoperability. For example, it may clarify how data processing is consistent with data protection policies, specifically for cross-agency "life-event" and "business-event" service chains across national and municipal authorities. The ministry and stakeholders should look to create a positive, accessible framework enabling data processing across siloes.
The Government Office, building on initial pilots and supported by adequate resource allocation, should scale secure, public-sector environments where public employees can experiment with administrative automation while ensuring data security and technical interoperability.
Launch a targeted strategic management programme for middle-tier civil servants (Government Office). Frontline staff are highly motivated to innovate with AI, but they face a "squeezed" middle management that lacks the strategic frameworks required to operationalise these ideas. The Government Office (in their role in developing management capability) needs to develop specific management modules focused on algorithmic procurement, process streamlining, and automation risk management. Upskilling middle management bridges the disconnect between leadership's operational focus and the frontline's digital ambitions, enabling the civil service to move from reactive administration to predictive governance. This gap is consistent with early evidence from the forthcoming OECD paper on AI skills (OECD, forthcoming[10]), which identifies building AI-capable leaders and middle managers as a distinct capability track, separate from technical upskilling, and one where most governments have made the least progress.
Create role-specific playbooks across all administrative tiers on AI adoption and experimentation (Ministry of Justice and Digital Affairs). Current generative AI adoption is uneven, driven primarily by managers who enjoy better resource access, while professional staff lack practical toolkits. There is a need for role-specific AI playbooks (such as "AI for Policy Analysts") and uniform experimentation guidelines to provide clear, immediate use cases that translate technical capability into daily public value. This asymmetry aligns with OECD evidence (OECD, forthcoming[10]) that highlights distinct needs across role types (from AI specialists to general civil servants) and project types (from off-the-shelf tools to custom models) and could offer Estonia a structure for developing these playbooks. These playbooks should for example delineate permissible data types for model prompting, standardised output validation protocols, and concrete, day-to-day use cases – such as summarizing public consultations, drafting policy briefs, or speeding up intake categorization.
Introduce service-wide "augmented administration" challenges to automate low-complexity, high-volume tasks (Ministry of Justice and Digital Affairs, Government Office and line ministries). Expand existing central pilot programs into service-wide "augmented administration" challenges to identify and automate low-complexity, high-volume tasks (Government Office, Ministry of Justice and Digital Affairs, and line ministries). Deploying advanced artificial intelligence tools directly onto legacy software and outdated procedures creates severe operational friction across local and central registries. Building directly on the Government Office's proven track record in facilitating central pilot programs, GO will drive the strategic orchestration and scaling of these challenges, while the Ministry of Justice and Digital Affairs provides horizontal technical guidance and reusable AI components. Through this central pilot mechanism, individual ministries can systematically audit their workflows to target repetitive administrative burdens suited for rapid algorithmic automation. Leveraging GO's existing piloting channels allows line ministries to test and prove these concepts in a controlled environment, securing quick, highly visible wins that reduce bureaucracy and demonstrate financial viability before scaling automation to complex, cross-agency service pipelines.
Implement cross-generational reverse mentoring schemes to standardise AI literacy across the public service (Ministry of Justice and Digital Affairs and all line ministries). The survey data reveals a sharp age divide, with public servants over the age of 60 reporting substantially lower AI literacy than colleagues under 40. Line ministries need to introduce structured, low-stakes reverse mentoring partnerships that pair junior "digital natives" with senior personnel. This formal knowledge transfer rapidly builds institutional confidence, ensuring that essential technical competencies and data literacy standards are shared equally across all age groups within the administration. Reverse mentoring of this kind is one example of the lateral capability-spreading mechanisms discussed in the forthcoming OECD paper on AI skills, alongside communities of practice and peer learning networks that allow skills to diffuse beyond specialist units (OECD, forthcoming[10]).
Restructure the funding model for AI public services to include mandated lifecycle maintenance costs and parallel operational funding (Ministry of Finance; Government Office and Ministry of Justice and Digital Affairs). Existing proactive digital services suffer from financial fragility because the state relies heavily on temporary project grants that fund only the initial development stage. Furthermore, transition costs are frequently underestimated; organisations need to run legacy systems in parallel with new AI-driven solutions (reportedly for a minimum of 6 to 24 months). The Ministry of Finance needs to transition these initiatives into the rolling state budget framework, requiring all public AI proposals to account for long-term maintenance, model drift validation, continuous data cleaning, and the necessary bridge funding to support dual-run operations. Assuring long-term fiscal sustainability prevents successful pilots from becoming isolated, non-interoperable experiments once grant allocations terminate.
Ensure sustainable AI capacity across central functions and policy domains to secure strategic continuity and execution leadership (Government Office, Ministry of Justice and Digital Affairs, and line ministries). The state's AI goals are constrained by limited specialist capacity, creating "key person risk" in agencies. The Government Office and the Ministry of Justice and Digital Affairs should secure flexible, sustainable funding models to reinforce central policy coordination, shared technical expertise, and implementation support. Rather than prescribing a rigid organizational setup, this central capability should leverage and align with ongoing IT consolidation across government. However, central capacity expansion must be balanced with targeted capability building within line ministries and agencies: strengthening central coordination must not dilute domain ownership; and sector-specific teams must retain sufficient operational capacity to ensure AI solutions remain directly grounded in domain-specific needs, data, and workflows.
Develop robust evaluation frameworks and practical guidance to demonstrate ROI on AI investments (Ministry of Finance, Government Office, Ministry of Justice and Digital Affairs). While existing national strategies mandate that artificial intelligence initiatives report against defined KPIs (Government of Estonia, 2024[6]), survey data reveals that actual evaluation capacity is missing or unpractised across the public sector (see also Chapter 6). Stakeholders underscored the importance of evaluating AI investments in terms of ROI to guide decision-makers. The Ministry of Finance, taking a leading role on financial methodologies alongside the Ministry of Justice and Digital Affairs and the Government Office, should issue evaluation guidelines on AI ROI, cost-benefit outcomes, and broader societal impacts to translate high-level strategy into everyday administrative practice. These guidelines need to provide practical and proportionate frameworks for combining quantitative metrics with qualitative case studies and annual independent reviews. In a proportionate manner depending on potential risks and level of impact, assessments need to explicitly measure human-rights impact, accessibility, interoperability, ROI and cost-benefit outcomes, and verifiable public value. Building this internal auditing capacity ensures that future deployment and scaling decisions are strictly grounded in empirical evidence rather than a superficial enthusiasm for the technology itself.
Establish structured forms of collaboration and joint proof-of-concept pipelines for university-public sector AI co-design (Government Office, Ministry of Justice and Digital Affairs, and Ministry of Education and Research). Rather than relying on informal networks, strategic AI projects should consider designating a named, accountable problem owner on both the public sector and university sides. Establishing clear, bilateral counterparts provides a human interface capable of proactively resolving intellectual property, data access, and procurement barriers before they stall development. Incorporating academic expertise early ensures that predictive public services are methodologically robust and publicly trusted before they reach active procurement.
References
[11] Finnish Government (2026), Estonia and Finland deepen cooperation on AI transformation, https://valtioneuvosto.fi/en/-/estonia-and-finland-deepen-cooperation-on-ai-transformation (accessed on 20 April 2026).
[3] Government of Estonia (2024), Andmete ja tehisintellekti valge raamat 2024-2030, https://e964029b-7c32-42a7-817c-37c391a3b976.filesusr.com/ugd/7df26f_fda04c9ce24e44f1ba33b17af3030b05.pdf (accessed on 13 July 2026).
[6] Government of Estonia (2024), Tehisintellekti tegevuskava., https://www.kratid.ee/_files/ugd/7df26f_21000a2dd36c4a66a30eea97563370a3.pdf.
[8] Government Office (2026), Eesti.ai algatus, https://riigikantselei.ee/valitsuse-too-planeerimine-ja-korraldamine/eestiai-algatus?_gl=1%2Ake462c%2A_ga%2AMzYxMTI2Mjk2LjE3NTkzOTQ4MTk.%2A_ga_E48CWYGN8L%2AczE3ODAzMDI1NDEkbzI5JGcxJHQxNzgwMzAyNjMxJGo2MCRsMCRoMA..
[9] Government Office (2026), Eesti.ai board gains three new members, https://www.riigikantselei.ee/en/news/eestiai-board-gains-three-new-members.
[7] Government Office (2026), Michal: Estonia aims to become a world leader in the use of artificial intelligence., https://www.riigikantselei.ee/en/news/michal-estonia-aims-become-world-leader-use-artificial-intelligence.
[4] Ministry of Economic Affairs and Communications (2021), Digital Agenda 2030, https://www.justdigi.ee/en/digital-communications-and-cyber/digital-agenda-2030.
[5] Ministry of Economics and Communication (2024), Andmete tegevuskava, https://www.kratid.ee/_files/ugd/7df26f_9796c76b908e492785242d909674a9bf.pdf.
[2] Ministry of Justice and Digital Affairs (2026), Eesti Digiühiskonna Arengukava, https://www.justdigi.ee/sites/default/files/documents/2026-05/D%C3%9CAK%202035%20arengukava.pdf.
[14] Ministry of Justice and Digital Affairs (2025), Tulevikukindla arvutusvõimekuse tagamine, https://www.justdigi.ee/sites/default/files/documents/2025-12/Tulevikukindla%20arvutusvo%CC%83imekuse%20tagamine%20v1.1_0.pdf.
[1] OECD (2026), Digital Government Outlook 2026: From Foundations to Transformational Impact, OECD Publishing, Paris, https://doi.org/10.1787/0496b2bc-en.
[17] OECD (2026), Generative AI experimentation in government: Learning from emerging guidelines, https://doi.org/10.1787/42815683-en.
[12] OECD (2026), Survey on Drivers of Trust in Public Institutions 2026 Results: Estonia, https://www.oecd.org/content/dam/oecd/en/publications/reports/2026/06/oecd-survey-on-drivers-of-trust-in-public-institutions-2026-results-country-notes_6dae3a76/estonia_d717c426/5e38c043-en.pdf (accessed on 13 July 2026).
[13] OECD (2025), Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, OECD Publishing, Paris, https://doi.org/10.1787/795de142-en.
[10] OECD (forthcoming), Bridging AI skills and talent gaps in the public sector: Governing with Artificial Intelligence.
[15] OECD (forthcoming), Governing with Agentic Artificial Intelligence.
[16] OECD (forthcoming), Governing with Artificial Intelligence: Measuring impact and demonstrating returns on AI investments.