How Slovenia's tax authority went from under 50% to 83% of callers reaching a human during peak season
800,000+
citizen questions answered, more than the call center itself
83%
callers reaching a human within 30-second SLA
5
specialized agents across tax and customs
At a glance
FURS is Slovenia's national tax and customs authority, serving 2.1 million citizens across five tax domains. For eight years, every question went through the same phone queue. Peak season fewer than half of callers reached a human within the service target. FURS needed a way to get routine answers to citizens instantly so human experts could focus on complex cases.
“Chatbots like ChatGPT and similar tools today offer a lot of general information, but it's not tailored to the life and business events of taxpayers. The FURS chatbot offers exactly that. It can identify the taxpayer's life or business event and adapt the information so it is comprehensive, complete, professional, and delivered in language that people who are not tax experts can understand.”
Tax experts trained the agents, not engineers
Two requirements shaped the platform choice. First, model flexibility. Tax content is unforgiving. A wrong answer about VAT deductions is not a service complaint, it is a compliance exposure. Being able to test different large language models and pick the best fit for each domain was a requirement, not a preference. Second, ownership. The people who knew the answers had to be the people who trained the agents. Any workflow that put an engineer in the middle of every content update would have created a bottleneck at the exact point where speed mattered most. FURS did not hire a technology team to interpret tax law. It taught tax experts to operate a chatbot and let their expertise flow straight into the agents.
Five specialized agents, each owned by a domain expert
FURS did not build one assistant that tries to know everything. It built five, each with its own domain and its own expert owner. Six of the most experienced contact-center staff became full-time bot trainers, people who had spent years answering the exact questions the agents were now taking on.
“The chatbot's operation is monitored daily by chatbot trainers. These are Financial Administration employees with experience working in the contact center, and they have poured that experience into the knowledge base. The trainers ensure the chatbot gets better every day. They review answers and upgrade the knowledge base so the chatbot keeps improving.”
The results
83% of callers reached a human within the 30-second service target during the 2026 peak tax season, up from under 50% the year before. The agents did not replace the phone line, they protected it. With routine questions absorbed, human staff were free to take the calls that needed them. 800,000+ citizen questions answered, more than the call center that used to carry all of them. The agents handle the volume that used to overwhelm the queue, consistently, around the clock. Human experts refocused on complex cases. Call center staff no longer spend their time on questions the agents can answer. The time goes to cases that require judgment, context, and follow-up. First place at Premik Naprej 2025, Slovenia's national award for innovation in public administration. The project received positive coverage in national tax and finance media immediately after launch.
What's next
The FURS team treats the chatbot as a living system, not a finished product. With five domains now live, the foundation is in place to expand the agents' role as the team identifies new opportunities.
“It makes our work at the Financial Administration much more modern and prepared for the future.”
More customer stories
How Aplazo doubled merchant acquisition without adding headcount
How Jumia J Force scaled support to 8 African markets with Chatbase
How West Coast Batteries solved complex battery recommendations with Chatbase
Better support starts today
Join thousands of teams using Chatbase to build AI agents that deliver exceptional support at scale — across chat, email, and voice.