A conversation with Kristel Kriisa, AI Advisor to Estonia's Secretary of State
Edition #111 The Estonian government's AI advisor reveals how a cautious and reflective approach has driven success in one of digital government's global leaders
For this piece, first published by aPolitical, I sat down with Kristel Kriisa - the AI Advisor to Estonia’s Secretary of State. Behind Estonia’s towering reputation for digital government, I found a team dedicated to creating change through reflection, learning and cautious experimentation. Early on in our discussion, Kriisa dismissed the idea of a “secret sauce” of Estonian innovation. The truth is much simpler - a commitment to digital transformation as a 30-year marathon, not a sprint.
Given Kristel Kriisa works in such a forward-looking field as AI, and in a country that recently became the first in the world to digitise 100% of its government services, our conversation focuses a lot on documenting and learning.
From one perspective, this is of little surprise. Estonia has been a global leader in digital transformation for decades now, and other countries are naturally often keen to learn the secret recipe for their success.
If they were looking for a drag-and-drop methodology, however, Kriisa makes it clear from the start of our conversation that they would be left disappointed. “It makes sense to go to places to see how they have done things, but you cannot just copy it”, she says. “People often don’t realise that it has taken Estonia decades to build it all up […] you cannot just adopt an approach of ‘give us the technology and in six months it will be done’”.
An Emerging Europe piece written to celebrate Estonia digitalizing 100% of its government services was equally quick to remind its readers that “this is the culmination of a 30-year journey from Soviet-era bureaucracy to a fully digital state”.
Earlier this year, Estonia became the first country to digitalize 100% of its bureaucracy.
And nor is Estonia too advanced to do its own learning. On a personal basis, Kriisa tells me she has a carefully curated Linkedin where she has “built a network of experts to give [her] updates on digital government around the world”. But nationally too, the story she tells me of Estonia’s approach to digital is founded on building its own institutional memory, and directing that approach to developing its services.
AI, knowledge management and building policy memory
This commitment to learning is perhaps best exhibited in an ongoing project to apply artificial intelligence to its knowledge management, an attempt to build better policy memory in government.
The potential for AI in this field is huge. Successive reports have shown that governments around the world are dogged by low institutional memory brought about by a high turnover of civil servants, the prevalence of contractors, regular restructuring of departments and poor IT management.
“The amount of data that we have in different ministries is crazy”, says Kriisa, “but often they are quite independent and owned by different agencies”. So “when you have a civil servant working on an issue, and they need to know everything that has been done before, it is really hard to find all of this information”. In this context, as I have written previously, AI can be used to build policy memory; unlock access to best practices in government; avoid duplication; and break down departmental silos.
To make sense of this data, Estonia “took 220k documents from across government – different strategies, regulations and research documents – cleaned them up and uploaded them to the knowledge base of an initial prototype to support semantic search and evidence-based policymaking.”. The first prototype was finished last August, followed by three months of testing to evaluate real-world performance and gather user feedback.
A second phase, currently ongoing, is attempting to improve the relevance and accuracy of search results. “The LLM is the easy bit”, says Kriisa, “the bigger challenge is having a good RAG system in place to pick the right documents, those that are relevant and give the best answer”. Kriisa and her team are currently working to refine those retrieval algorithms to better select the most relevant segments, finding a balance between information overload and completeness.
On top of that, ensuring that the platform’s “knowledge base stays comprehensive and up-to-date is also one of the priorities”. Alongside the RAG refinement, the government is working on dynamic knowledge base expansion. Practically, this means integrating new documents while attempting to maintain metadata quality and consistency, all while ensuring high security and accessibility standards.
“Cautious and reflective”: The Estonian approach to piloting and scaling.
It is clear the knowledge management system is work in progress, but Kriisa and her team will be encouraged by the track record of Estonian success.
Much of the approach that has allowed Estonia to become the first country in the world to digitise 100% of its government services is founded on piloting and scaling. But even that, as Kriisa tells me, is a mindset that has been subject to learning and iteration.
“When we first started implementing AI in the public sector back in 2018, we thought the best way was to experiment as much as possible, as that was what other countries were doing”, Kriisa tells me. Setting the country up for success with these pilots involved practically learning from failures and iterating quickly - “being comfortable with the idea of investing in learning experiences, even if not every project delivered immediate results”.
It has also involved learning to prioritise what challenge the government decides to work on. “When you have limited resources like Estonia, this can be very hard”, says Kriisa. “Sometimes you see a solution that would help so many people, but there are regulations in the way that mean it would just take so much time”.
Or, to take another example that relates to the knowledge management platform, Kriisa admits to being tempted to think about building an Estonian sovereign LLM a couple of years ago, especially seeing peers like the Netherlands or Kazakhstan investing in doing so. Now, “we have realized that doesn’t make so much sense in such a small country, given the need for so much money, infrastructure and experts”.
What this has meant in practice is learning to prioritize more before experimenting. The same approach to testing ideas quickly and with small amounts of money has remained, but “we have realised we have to pick our battles […] try to find pain points and see where we can actually change things considerably”.
This approach is clearly defined in the 3rd iteration of Estonia’s AI strategy, themed around “collaboration, reusability and infrastructure”. Building things modular, open-source and publicly has allowed government departments to reuse components, and to make their limited resources go further. It is the same principle as the knowledge management software – making ideas and technology go further.
Kriisa describes her journey developing Estonia’s digital government as one “that has required patience, resilience, and strategic reflection”.
For a country with such an innovative reputation as Estonia, this approach of stopping and reflecting may be initially surprising, far from the “move fast and break things” approach that is culturally associated with innovation. But, to return to Kriisa’s advice to would-be partner countries of learning versus copying, it is evidence that digital transformation is a 30-year marathon, not a sprint.