In May, Estonia’s latest member of the workforce will start work proving personalised support to citizens when they use government services.
Named Bürokratt, the Siri-like digital assistant will help citizens carry out tasks such as submitting applications and making payments, all through voice commands.
It will even proactively contact citizens to remind them if they need to complete a task such as renew a driving licence, register a birth or file their tax return, and promises to save the government millions of Euros in administrative costs to make citizens’ lives easier.
There is no other project like it, and it’s another example of Estonia leading the way in digital government.
Ott Velsberg, chief digital officer of Estonia’s ministry of economic affairs and communications, said that when he joined the ministry in 2018, his first public sector role, he began working on how artificial intelligence (AI) could revolutionise government, education and the private sector.
Velsberg, who has a PhD in informatics, had previously only worked in the private sector focused on the internet of things (IoT) and data analytics with a part-time university lecturing role.
“One of the first things I started working on was forming the strategy of how we take up AI in the government, in the private sector, and how it can boost education and research,” said Velsberg.
AI strategy
In 2019, he and his team came up with an AI strategy, which included a concept of what Bürokratt could be. “By the end of 2020, we came up with that very concept,” said Velsberg.
“The first 18 months was about validating if what we were working on would be beneficial, and could work through proof of concepts and minimal viable products,” he said.
“As far as we know, no other government or public sector organisation has been doing anything similar to Bürokratt.”
Building the team
At first, the team was made up of existing department staff including Velsberg, but he soon needed to add specialisation. “Initially it was just for the proof of concepts, but last year, when most were completed and we saw that there was tremendous potential in what we were working on, I began to form a core team.”
The team of 13 was built with skills from the private sector, with just three of the members from the public sector. “We were looking for people from outside because it is really about the skills and, in government, there is typically not that much in terms of the data analytics skills needed, nor a similar project.”
Architects, developers that are now project managers, data scientists, people that have worked specifically in language technology and a business analyst make up the mixed group, said Velsberg.
To retain support from the government, it was necessary from the first day to show Bürokratt could increase efficiency, provide better services to citizens and improve accessibility.
An efficiency perspective
Velsberg said that in a medium to large-sized enterprise, just forwarding emails takes the annual work time of six people, with the average cost of responding to a citizen request through a person ranging between €4 and €12.
“If we can automate just 30% of email forwarding and extend it to different services, we could save €1m a year,” said Velsberg.
He said the project itself to create the platform costs a little under €500,000, which he said was much lower than what licences would cost for off-the-shelf chatbots across the government departments interested for just one year.
“There was interest from 20 agencies to introduce a chatbot, and the development costs for the internal platform was less than the one year license costs for third-party chatbots,” said Velsberg.
The first step was building the base platform, which has user authentication at its core. Since then, the team has been adding functionality, some of which is expected to go live in May this year. The first service to go live will be personalised weather forecasts.
“This is a safe and foolproof solution,” said Velsberg. “We don’t want to go live with a service ordering documents because we want to keep the risk lower.
“We already have public APIs [application programming interfaces] we can use, and it’s about people receiving weather information when they want it based on where they are,” he said.
But the plans are far greater and the project is continuous, with an initial roadmap until 2025. “We want most major agencies to be part of Bürokratt with the government functioning as one channel 24 hours a day,” said Velsberg.
However, there are challenges to face before hitting this target, including getting the speech recognition right.
“One of the most difficult parts is getting our language technology up there with the technology that exists today in English, for instance,” he said.
To this end, the government plans a nationwide programme known as Donate Your Speech, which aims to get at least 4,000 hours of speech data this year to help improve voice recognition. “By 2025, we want our speech recognition a 95%, which is on a par with human recognition,” said Velsberg.
Supported languages
The first languages to be supported are Estonian, Russian and English.
“We will then add more languages if there is a need,” he said, providing an example of automating the process of fishermen getting permission to fish in Estonian waters: “A lot of fishermen are Latvian, so we are considering adding the language.”
Another challenge arises from scale. Estonia is a small country with just over 1.3 million citizens, so it can in many instances be difficult to train models to work due to a lack of data. “Many agencies simply don’t receive many requests,” said Velsberg.
Estonia already has a reputation for state-of-the-art digital government, and Bürokratt is another first, so the world will be watching.
“I already see interest from other countries, and many have approached us,” he said. “They are not just small countries, and some are outside of Europe. The solution is open source, so everyone can use it.”
