Estonia’s answer to the future of customer support could help revolutionise the industry and offer significant value for operators – but there are still plenty of challenges to consider.

Based in Estonia’s capital, Tallinn, the company is pitching itself as an infrastructure-level partner for betting operators and hoping to revolutionise how customer support thinks and acts in the present day. 

In exclusive comments to SiGMA News, summing up the changing role of support, Harpo Lilja is clear: “Support isn’t just about answers. It’s about actions that match intent, brand, and regulation.”

There are many benefits to AI; defence against regulatory breaches, reputational damage, and player churn, for example. However, in the gambling industry, patience can run thin as emotions run high for customers. 

That’s where a more considered approach comes in. The focus has been on the changing role of support for the Estonia-based company. 

In terms of Tugi Tark, which translates as smart support, trained its AI agents on more than 10 million resolved iGaming tickets. Following its launch, it supports over 250 languages but there is a focus on accuracy rather than speed. 

“The current model is focused on a smaller subset,” Lilja explains. “We use multiple language models, especially slower, more accurate ones, to refine emotional accuracy. Ultimately, the quality of interaction depends on the quality of the training data.”

What AI can’t do yet

In terms of the AI’s limitations, it certainly can’t do everything from a support point of view, yet. With high-stakes queries, disputes, emotional stress, and VIP escalation being difficult behaviours to replicate.

Players often are asking to speak to a human and they are routed according to a basic ruleset, but there are concrete plans to expand this functionality as the platform scale. “Our AI agent knows both the player data and brand configuration, so customising this behaviour is trivial,” Lilja says.

One question to be asked is how much of the system is truly AI-driven versus rule-based logic? On that topic, Lilja broke it down: “It’s not an either/or situation. Rules are guardrails. LLMs are the only way to process and create human responses at scale, but they require constraints.

“Around 70 percent of the agent’s functionality is pure AI, ranging from ticket categorisation to complex reasoning tailored to a customer’s needs.”