Evaluating Public Sector AI Adoption: A Philippines Case Study
Evaluating Public Sector AI Adoption: A Philippines Case Study
A practical EvalCommunity Academy case study for evaluators, M&E specialists, researchers, and development practitioners working on AI, governance, digital transformation, and public sector reform.
Last updated: May 19, 2026 · 8 min read · 1,470 words
Introduction
Public sector AI adoption refers to the way government agencies design, develop, deploy, manage, and sustain artificial intelligence systems for public services, administration, decision-making, and citizen engagement. For evaluators, the issue is not only whether an AI tool works technically. The more important question is whether the system is institutionally ready, ethically governed, financially sustainable, and useful in real public sector conditions.
This EvalCommunity Academy case study is based on the peer-reviewed article Lessons from Public Sector AI Adoption in Developing Countries: An AI Life Cycle Perspective. The study examines seven AI initiatives in the Philippine public sector using semi-structured interviews with ten stakeholders. It analyzes the cases across three AI life cycle stages: design, development, and deployment.
The article finds that strong leadership helped AI initiatives begin, but later stages were constrained by infrastructure gaps, financial instability, talent attrition, limited data readiness, and weak governance arrangements. These findings are highly relevant for evaluators and M&E professionals who assess AI, digital transformation, and public sector reform projects in developing-country and Global South contexts.
Quick Answer
Public sector AI adoption succeeds when governments evaluate more than technology. Evaluators should assess data readiness, infrastructure, talent retention, governance, funding, user acceptance, and the pathway from pilot to sustainable deployment.
Key Takeaways
- AI projects in government should be evaluated as socio-technical systems, not only as software tools.
- Strong leadership can start AI projects, but it does not guarantee sustainability.
- Data readiness, infrastructure, and computing capacity shape what is possible from the design stage.
- Talent attrition is a major risk when trained public sector staff move to higher-paying private sector roles.
- Many AI pilots stall when there is no implementing agency, deployment funding, or scaling roadmap.
- Localized AI governance is needed for accountability, privacy, maintenance, and responsible use.
Table of Contents
Case Background
The Philippines provides a useful setting for examining public sector AI adoption in developing countries. The article describes the country as facing digital divide challenges, inadequate infrastructure, digital literacy gaps, cybersecurity concerns, and limited resources, while also making progress in digital government and AI readiness.
The study examined public service organizations across different administrative levels. It used three analytical lenses: the Technology-Organization-Environment framework, Technology Affordances and Constraints Theory, and an AI life cycle perspective. Together, these lenses helped explain why AI projects are adopted, how they are used, and where they encounter constraints across design, development, and deployment.
The Seven AI Cases
The article anonymized the projects as Cases A to G. Four cases were at the national government level, two were at the local government level, and one was within a state university. Most were internally developed, except for two local-level initiatives.
- Case A: A city-level local government computer vision initiative, outsourced by the city government, deployed and operational at the time of writing.
- Case B: A barangay-level proof-of-concept computer vision initiative, deployed during COVID-19 and later discontinued.
- Case C: A national government computer vision initiative serving internal and external users, still under development.
- Case D: A national government natural language processing initiative, including work on local-language datasets.
- Case E: A national government mobile application combining chatbot functionality and predictive analytics for citizens.
- Case F: A national government computer vision initiative still training its AI model, with deployment not covered by current funding.
- Case G: A state university computer vision initiative with prototype demonstrations but no implementing agency to operationalize it.
The Evaluation Problem in Public Sector AI Adoption
The central evaluation problem is that AI pilots can look promising while remaining institutionally fragile. A model may work in a demonstration, but still fail to become a sustainable public service system if the agency lacks infrastructure, staff, governance, funding, or operational ownership.
For evaluators and M&E professionals, this means public sector AI adoption should be assessed across the full life cycle. Design-stage readiness, development-stage capacity, and deployment-stage sustainability all require different evaluation questions.
Main Findings for Evaluators
1. Leadership helps start AI projects, but does not guarantee survival
Across the cases, top management support helped align AI projects with organizational goals, secure support, and move ideas into early implementation. However, leadership did not remove structural constraints. Projects still depended on data readiness, digital infrastructure, budgets, and institutional continuity.
2. Infrastructure and data readiness shape the design stage
Cases with existing command centers, CCTV systems, or digital platforms were better positioned to explore AI. Still, additional investments such as fiber-optic networks, cloud infrastructure, data centers, and high-performance computing were often needed. Fragmented, unstructured, or inconsistently governed data limited what agencies could realistically design.
3. Talent retention is a major development bottleneck
The study found that public organizations often lacked permanent AI or IT roles and relied on contractual staff or fresh graduates. When trained staff left for better-paying private sector roles, agencies lost technical knowledge and had to rebuild capacity. This made development slower and more fragile.
4. Deployment requires governance, funding, and ownership
Several projects reached deployment, but others remained in development or prototype stages. Case G had completed AI development and demonstrations but lacked an implementing agency. Case F was still training its model, but deployment was not included in the current funding allocation. These examples show why evaluators must assess the transition from pilot to institutional use.
5. User acceptance and accountability matter after launch
Deployment introduced practical challenges such as bugs, downtime, maintenance, security patches, hardware upkeep, workload changes, and accountability for errors. In one case, concerns about job displacement reduced user acceptance. This highlights the need for communication, change management, and clear governance arrangements.
Evaluation Framework for AI Projects
EvalCommunity Academy users can adapt the following framework when reviewing public sector AI projects.
Design-stage questions
- Is AI necessary, or would a simpler digital solution solve the problem?
- Are relevant datasets available, usable, secure, and governed?
- Does the agency have the infrastructure to support AI development?
- Are privacy, accessibility, bias, and security risks identified early?
- Is the project aligned with the agency’s mandate and public value goals?
Development-stage questions
- Does the agency have the technical and domain expertise needed?
- Are staff skills retained within the institution?
- Is the AI system adapted to local language, workflows, and administrative realities?
- Can the system integrate with existing databases and legacy systems?
- Are model assumptions, data processes, and testing results documented?
Deployment-stage questions
- Who owns the system after launch?
- Is there funding for maintenance, updates, cybersecurity, and scaling?
- Who is accountable for AI-related errors or harms?
- Are users trained and willing to adopt the system?
- Is there a realistic roadmap from pilot to routine public service delivery?
Practical Lessons for M&E and Development Professionals
First, evaluate AI as an institutional change process. Technical outputs matter, but they are not enough. The evaluation should also examine people, workflows, governance, incentives, and long-term ownership.
Second, do not confuse prototype success with public value. A prototype can be promising while still lacking funding, operational ownership, user acceptance, or a deployment pathway.
Third, include sustainability from the beginning. The article’s findings show that talent retention, infrastructure investment, and localized governance are essential for moving beyond pilots.
Finally, evaluators should include frontline staff and citizens when possible. The article notes that its evidence primarily reflects project leads and senior officials, which means future evaluations should add user and citizen perspectives.
Download the Original Article
Keep the original peer-reviewed article as a reference for your evaluation work, M&E planning, AI governance analysis, or professional development.
Useful External Resources
- UN E-Government Survey 2024 for digital government context and benchmarking.
- UNESCO Recommendation on the Ethics of Artificial Intelligence for responsible AI principles.
- OECD AI Principles for trustworthy and human-centered AI governance.
- UNDP Digital Strategy 2022–2025 for inclusive digital transformation in development contexts.
FAQ
What is public sector AI adoption?
Public sector AI adoption is the process through which government agencies design, develop, deploy, and manage artificial intelligence systems for public administration and service delivery. It includes technology, governance, staffing, data, funding, and user acceptance.
Why is AI adoption difficult in developing countries?
The case shows that developing-country agencies may face infrastructure gaps, fragmented data, limited budgets, talent attrition, and weak localized governance. These constraints become especially visible during development and deployment.
What should evaluators examine before an AI pilot scales?
Evaluators should examine data readiness, infrastructure, staff capacity, governance, user acceptance, accountability, maintenance funding, and institutional ownership. A technically successful pilot is not automatically ready for scale.
What was the biggest challenge in the Philippines case?
The article identifies several major challenges, including limited infrastructure, financial instability, talent attrition, and weak governance. Talent retention was especially important because trained staff often moved to higher-paying private sector roles.
How can AI evaluation support better public service delivery?
AI evaluation can help agencies identify whether a system is useful, safe, sustainable, and aligned with public value. It can also reveal whether the organization is ready to maintain and govern the system after deployment.
Why does the AI life cycle matter for M&E?
The AI life cycle helps evaluators separate design, development, and deployment risks. Each stage has different evidence needs, from problem definition and data readiness to model testing, user adoption, monitoring, and sustainability.
Conclusion
This case study shows that public sector AI adoption is not simply a question of technology. In the Philippine cases, AI initiatives were shaped by leadership, infrastructure, data readiness, staff capacity, financing, governance, user acceptance, and institutional ownership.
For EvalCommunity Academy users, the main lesson is practical: evaluate AI projects across the full life cycle. A responsible evaluation should ask whether the system is technically feasible, institutionally sustainable, ethically governed, financially supported, and capable of delivering public value beyond the pilot phase.
