An AI lesson from Estonia

In Estonia, AI is known as Kratt, after a mythological creature that does everything its master asks of it. And it is living up to its name.

This tiny Baltic nation, a pioneer in digital government (where everything from voting to getting divorced can be done online), adopted its first AI strategy in 2019.

“Currently, there are over 200 identified AI use cases across Estonia’s central government,” says Ott Velsberg, Estonia’s government chief data officer. And with 37% of public servants applying AI solutions in their roles, Kratt is well and truly entrenched in processes and service delivery.

Its most successful uses include AI solutions implemented by the Estonian Unemployment Insurance Fund to link jobseekers with relevant vacancies and identify people at risk of long-term unemployment. “This has not only reduced the workload for case officers but also shortened the time people spend unemployed,” Velsberg says.

Estonia also uses AI-driven tax fraud detection systems, automatic transcription of court hearings and parliament sessions, real-time subtitling on national television, transport robots in hospitals and autonomous vehicles.

It also has “Bürokratt”. To Estonian citizens, this is a chatbot that links them to government services. To Velsberg, it’s much more than that.

“Bürokratt is not one bot,” he says. “It’s a national architecture for conversational services.”

Bürokratt connects large language models (LLMs) to official services under public sector rules, Velsberg says.

“It handles secure identity, consent, routing, audit, and integrations with government back-ends. It can orchestrate multiple models – including commercial LLMs – behind policy and logging layers, swapping providers as needed.”

Fifteen government organisations use it, with 18 working on adoption.

Shared tools and lots of support

Estonia’s approach to scaling AI across government has been twofold, Velsberg says. It offers targeted support and shared tools and services.

An AI support toolbox provides practical support for agencies and organisations developing or implementing AI solutions. This includes deep-dive assessments, advisory services, and guidance on procurement, helping institutions make informed choices, manage risks, and design AI that is fit for purpose.

“The toolbox ensures that even smaller agencies with limited in-house expertise can still adopt AI responsibly and effectively,” Velsberg says.

The second part of the approach is providing core tools and services that can be used in different contexts. “We have invested in a set of shared AI tools, platforms, and services that are accessible across government and beyond,” he says.

“These include reusable components, frameworks, and infrastructure that can be applied in different contexts. By providing these as common goods, we help avoid duplicative investments, accelerate adoption, and create consistency across the ecosystem.”

Challenges and issues

The main challenge Estonia has faced on this journey has been a shortage of knowledge and skills, Velsberg says. As a result, the nation is investing heavily in training programs for data stewards, AI specialists and public servants.

One of the biggest lessons Estonia has learnt is that moving from pilots to production requires more than just good models. “Success depends on building the right enablers and governance structures around AI,” Velsberg says. “Pilots are often funded as experiments, but scaling to production means ensuring long-term financing, clear ownership, and predictable maintenance costs. Without dedicated budgets, even the most promising AI solutions risk stalling.”

Service integration has also been a challenge; AI cannot simply be bolted on to existing services, Velsberg says. “Processes often need to be rethought end-to-end to take full advantage of automation and predictive capabilities. This means treating AI adoption as an opportunity to reimagine service delivery, not just to digitise the old.”

The main strategic focus now for Estonia is the AI Leap initiative, which focuses on two areas: education and public administration. On the education side, students and teachers are to get access to new AI-powered educational tools, and teachers will receive systematic training in using them. On the public administration side, by 2026 Estonia aims to equip about 10,800 public servants with AI-based tools that will enhance efficiency, automate routine processes, and strengthen evidence-based decision-making.

“By scaling these solutions across government, our goal is to save roughly 21 million hours of administrative work annually by 2030, freeing up resources for higher-value tasks and improving the citizen experience,” Velsberg says.

Is Australia ready?

Estonia’s integration of AI has been impressive. But Kersti Eesmaa, the managing director of WorkPath Australia, a board member at the Australian Cyber Collaboration Centre, and the former Estonian ambassador to Australia, says it’s important to realise that it didn’t take place in a vacuum.

“We didn’t start this from scratch,” she says. “We already had two layers. Firstly, the digital layer: Estonia has been a digital society for 20 years. And secondly, cybersecurity. For AI to be successful in Australian government or society, these two layers need to be there.”

As such, Australia finds itself in a tricky position, Eesmaa says. It can’t afford not to embrace AI, yet it is going to struggle to implement efficient AI systems because its cybersecurity is not as robust as Estonia’s, and its digital services are not connected.

“There is no interoperability, and more importantly, there is a challenge with data, because for AI to function really well in government, you need to have good-quality data.

“I can’t think of a successful AI model that didn’t have digital government rolled out really well first. It’s like trying to build the roof without having a foundation or the walls up.”

How to tackle this is a big question, Eesmaa says, and it’s understandable that Australia is taking its time to work through the issues. But governments should be wary of being too risk-averse

“There’s a risk that Australia will just get stuck admiring the problem, which we can’t afford to do, because everybody else in the world is moving. My point is: let’s not jump on the bandwagon while forgetting those other layers. Australia needs to be aware that it also has to work on those foundations.”