The Estonian Investment Agency (Invest Estonia) has twice been included in IRCAI’s (The International Research Centre in Artificial Intelligence under the auspices of UNESCO) list of 100 most promising artificial intelligence related solutions for the benefit of humanity. One of the agency’s solutions partly relying on machine learning, ComparEST, was recently shortlisted in Emerging Europe Awards as the Policymaking Initiative of the Year. In 2020, The Estonian Investment Agency was awarded the United Nations Investment Promotion Award as, according to the U.N, the agency has shown readiness, innovation and excellence in supporting the evolving needs of investors during the COVID-19 crisis – this involved implementing Suve, the AI-based chatbot currently in use on Invest Estonia’s webpage, as a country-wide tool to spread information during the COVID-19 related emergency situation.
How has Invest Estonia utilized AI for investment promotion?
A.K: First of all – for us the goal has never been ‘using AI’ – it has always been providing the best possible service to our customers as efficiently as possible. Automation, including the use of AI is just a way to achieve this.
K.V: Still, we can consider the use of AI at our agency as a kind of success story – for two years in a row, automation at the Estonian Investment Agency has been included in UNESCO’s list of top 100 projects solving problems related to the United Nations Sustainable Development Goals. And what we have also done – saved tax payers’ money and helped more potential investors than we could ever have helped manually.
A.K: We have employed a set of what we call ‘non-human agents’ or ‘robotic colleagues’ working at our investment promotion agency. Suve is most probably the easiest tool to grasp – a chatbot who is run in cooperation with our colleagues at Work in Estonia, the agency attracting new talent to the country. We have seen COVID-19 influence a lot of IPAs around the world and she is what we could call our COVID success story. Our agency’s employees were a part of a team of volunteers who first deployed the same technology in the toughest of possible environments – as a health crisis communication tool for the government. After the health situation improved, it was obvious that the crisis time tool could also have a ‘peace-time’ application at the agency. We also know how to use Suve in crisis communication, which is a big plus, though we hope we will not need to do it.
K.V: Eia is our oldest and this far most productive robotic colleague – leads coming from our webpage first go to Eia who sets service levels and then replies to enquiries automatically or semi-automatically, including quite elaborate investment offers in the replies. Here, statistics combined with quite simple but efficient machine learning models is used, besides obviously a lot of conditional logic. Eia is one of the few robots in Estonia who has done a successful marketing campaign with our President Kersti Kaljulaid.
And Emma is our media monitoring and social media handling assistant. She uses several machine learning models in media monitoring and also has access to what is the trend today – a large language model (LLM). The reasons for limiting OpenAI use but still having a try are currently obvious and universal – possible confidentiality concerns and quality assurance. We are experimenting with multiple LLM-based solutions for improving both our internal processes in the agency and also smoothening the client journey. But as we are only in the testing phase it is still too early to talk about them.
What tools have you found useful?
A.K: For us, the question is not about what particular tools or technologies work – the possibilities change constantly and we just have to adopt. The real question is what the universal success factors of automation at our agency are. Systematic research into this, as well as gathering best practices across countries and industries, is where we have heavily invested our time. While there are technological success factors, such as ease of ensuring information security or allowing integrations with other services, most of the success factors we have identified, are actually human and communications related. For example, our digital processes have to go hand in hand with what is happening offline at the agency – and ensuring this can sometimes be a challenge.
K.V: Talking about particular tools – what we see a lot of potential in, is using simple statistics, as well as building our own machine learning models that solve quite narrow questions but do it at quality that is similar to a specialist solving them.
What challenges have you encountered with them?
K.V: As Andero often says, technology is simple, it is communications and people that are difficult. Of course, we have had our technological challenges, as well. For example, there is always the question if, for a certain task, custom-made software is needed or we can use out-of-the-box software and simply integrate it into our toolset.
A.K: As much as I have heard, this is the dilemma at many investment agencies across the world. A few times, we have chosen a solution with not enough possibilities to handle our exceptions or integrate with other tools, and the initial plan has failed because of that. But this does not mean that we wouldn’t learn and try again.
K.V: Another question is being able to ensure that a tool is needed and will be used. We have to take into account that usability is a function of usefulness and ease of use, so things need to be kept simple to use but at the same time sophisticated – which is always a challenge.
How do you address the question of AI ethics?
A.K: When using AI, there are always ethical concerns. We take them seriously and when planning major process updates, have involved independent parties in the planning process to have a second opinion on legality and possible moral aspects of what is being planned.
How do you make the agency’s employees view AI as an opportunity, not a threat?
K.V: What we have found very useful is having your internal KPIs and dashboard presenting the time saved by automation. It helps the whole team understand why you should constantly contribute your resources to development instead of following the same patterns, nor keep doing things the old way just because it’s a convenient habit. And when introducing the new tools, we have found the communication part crucial for adaption. That is also the reason why we have names and personas for our robots, and we talk about them as our restless colleagues, who are happy to work 24/7, never taking a vacation and always being in the same good mood, like we have taught them.
Wish to get 1:1 e-Consulting from one of our investment advisors, helped by Eia, the electronic investment advisor? Click here to get started.