Podcast 🎧 & blog: The learning curve for effective AI in government
Governments worldwide are eager to harness AI — yet many remain stronger on ambition than readiness. In this episode, “The Learning Curve for Effective AI in Government,” we talk with Piret Hirv, Head of the Data Management Competence Centre at the e-Governance Academy, about what it takes to turn AI pilots into lasting impact in public services.
Written by Federico Plantera
The latest OECD report Governing with Artificial Intelligence presented in September, finds that many governments are running on ambition, but short on readiness. There is no lack of strategies, pilot projects, and hype, but far fewer examples of AI being scaled and embedded into everyday governance. The tension between ambitious plans and adoption-to-implementation realities is where our conversation with Piret Hirv starts, as Head of the Data Management Competence Centre at e-Governance Academy.
“Implementing AI is way beyond technology or regulation. It is about people, processes, and clarity on goals and tools. First, you need to ask: what problem are we solving, and what function can AI actually support or replace?” Hirv stresses. Without that framing, pilots risk being clever experiments with little impact.
The OECD report highlighted the same pattern: experimentation without scale. While pilots are essential, governments often face familiar obstacles when trying to move beyond them. These obstacles include fragmented data, siloed systems, limited in-house expertise, and governance structures that lag behind technological progress.
Cutting Through the Hype
Governments, as the OECD observed, are approaching AI with enthusiasm but uneven capacity. Hirv noted that enthusiasm alone can be misleading: “You cannot build resilience on motivation alone. You need the architecture and the practice to make it real.” AI should never be a shiny object, but rather a tool that supports existing processes – ones that must be redesigned and clarified first. Otherwise, AI risks simply masking inefficiencies rather than solving them.
Public authorities in Spain and Germany, who work with us and the EU to bring advanced digital tools to entrepreneurial services, have seen the same reality firsthand. As Oscar Del Campo Barxias, Deputy Director General for Entrepreneurship of the Community of Madrid, explains, “For us, the implementation of artificial intelligence in our region represents a major step forward in how we design and deliver public services for entrepreneurs. Our vision is not based on replacing, but on using AI as a tool to help our staff and citizens.”
“Through AI data analysis, we can better understand the needs of startups and small businesses, anticipating challenges and offering a better service. AI enables a more dynamic and inclusive entrepreneurial ecosystem,” he adds.
Interest in AI is strong, but practical foundations often determine whether progress can be sustained. Poor data quality, absent interoperability, or unclear ownership of services – all risk hindering genuine intention and willingness to improve. Not glamorous obstacles, but still, decisive ones.
Enablers Before Technology
The OECD report outlines a set of enablers governments must have in place before AI can work in practice. Hirv broke them down succinctly: data, processes, governance, skills, interoperability. Without them, AI remains a patch on dysfunctional systems.
“If you haven’t sorted out your processes in service delivery, then what you get is a messy process, smoothly covered with AI,” Hirv explains. The Madrid region’s experience echoes this. As Del Campo Barxias notes, “AI projects must begin with a clear purpose. Because AI often requires sharing critical data, it links directly with transparency. So you must also be able to explain how algorithms work, for citizens and personnel alike to feel confident and safe about the process.”
Here are some of the OECD’s key findings on this discussion:
- Scaling up is the main challenge
Many AI projects remain stuck at the pilot stage, unable to move into regular service delivery. - Poor data quality undermines AI efforts
Governments still struggle with fragmented datasets, siloed systems, and the absence of interoperability. - Skills gaps are decisive
AI skills age quickly, and without organisational capacity, governments risk perpetual dependence on external providers. - Return on investment is unclear
Very few projects systematically calculate cost savings or societal value, which risks undermining confidence. - Guardrails are emerging unevenly
Some countries use binding principles, others non-binding guidelines, but trustworthiness remains central to adoption.
Capacity That Lasts
Beyond processes and tools, capacity building emerged as one of the most pressing concerns. Too often, governments see training as a one-off event. But as Hirv points out, “the only solution is to train your own experts and motivate them to remain in government. External expertise can help, but only if it leaves knowledge behind.” Skills in AI age quickly, so the challenge is not just recruiting talent but embedding organisational memory – structures that preserve knowledge and keep it accessible as technologies evolve.
This is why peer learning, cross-border exchanges, and shared platforms matter so much. They help institutions maintain capacity even when individual staff move on. OECD evidence supports this: administrations that view AI adoption as a procurement task tend to struggle, while those that invest in communities of practice and continuous process improvement sustain resilience over time.
Building confidence in results can fruitfully contribute to this, and to scaling AI past pilots. The OECD warns that governments rarely calculate return on investment for AI projects, whether in financial or societal terms.
This is risky. Pilots, instead, should be designed with users from the start: “Even prototypes must be designed together with users. If you don’t fix real-life problems, uptake will be poor.” Benefits may not always appear on balance sheets, but smoother company registration, reduced fraud, or faster service delivery are all important markers of success.
It will take a certain degree of experimentation, one that must also come with tolerance for failure. “Innovation can fail. That is fine. The key is to make lessons affordable and to ensure they are learned,” Hirv reassures. Suppose AI strategies are to survive political cycles and deliver value over the long term. In that case, policymakers must shift from a mentality focused on avoiding risk to one of managing it consciously.
The Learning Curve, Beyond Chatbots
Lastly, practitioners and enthusiasts often equate AI with a single tool: chatbots. As Hirv noted, this view risks narrowing too much on mainstream functionality, missing the diversity of possible use cases. “Governments should not end up with 30 chatbots for 30 different services. The real goal is to redesign processes and embed AI where it genuinely helps.”
In Estonia, for example, there are nearly 200 AI-based solutions already in use in government – but only a fraction are visible to citizens. Many function in the background: fraud detection, risk modelling, workflow optimisation. The lesson is that AI is not a single application but a set of instruments, and organisations need to learn how to match tools to real problems.
There is an institutional learning curve because AI in government is neither something to be rushed into nor abandoned when difficulties arise. And adoption comes when organisations develop the capacity to evaluate and adapt AI over time.
If we see AI as a silver bullet, the risk is to just bury the nodes and links in service process improvement that require attention in the very first place. But when embedded thoughtfully, it can help governments simplify and deliver, matching the needs of citizens and businesses alike.
As Hirv reminds us: “Digital transformation is never cheap. But the benefits may be elsewhere – in faster services, in citizens’ trust, in opportunities for entrepreneurs. That is why governments must keep moving, step by step, learning as they go.”
The podcast episode is part of the EU-supported project  “Supporting regional entrepreneurship through the adoption of innovative technologies, including AI, in public services”. The project aims to create a foundation for the next generation of disruptive services for entrepreneurs in Germany and Spain by integrating innovative technologies, including AI.