Trust. Not a word commonly associated with technology, but something that carries significant weight when referring to AI. For iGaming technology firm SOFTSWISS’s recently appointed Chief AI Officer (CAIO), Denis Romanovskiy, employee trust in the tools and processes creates the framework for improved workflow, leading to better customer support and polished products. But trust is a process.
Romanovskiy admits that he was initially “quite skeptical about AI”, and only when models began to improve, “particular cases were automated, and I myself started to use these tools day after day”, did he start to believe that “this is a very big, very complex, absolutely new era”.
To maximize its potential for SOFTWSWISS, this required someone dedicating their full focus to AI, but the most interesting part of this was the human element – not just overcoming skepticism but also encouraging experimentation and truly understanding how people work.
This involves weaving together the proper elements – with the company using tools from Google, Anthropic, OpenAI, Oracle and other vendors – with the proper mindset, enabling employees with “very simple cases” such as creating emails or summarizing data, and then enabling “your corporate system to work together with the AI systems”.
This has to be done in a “safe, careful way”, even individually showing employees how the process can work and watching their fascination when they realize “Yes, they can do it themselves”. Once they’re confident, then you can “enable them with corporate data, and then you need to build some of your own infrastructure tools to enable all of these components” – oftentimes referred to as “agentic workflow”.
Building is easy, knowledge is hard
For internal workflows, “everybody is now adopting” AI. But that doesn’t worry the CAIO, as the possibility of SOFTSWISS losing its uniqueness is “not a sudden risk. It’s more of a long-term risk”.
“Building something is easy, now it’s very easy. But you need to understand exactly what to build, how to build it, how to test it for quality, how to test it in production, and how to integrate it into the work of your clients. And that knowledge is not that easy to gain. You need years for that.”
“If you do not move and adopt the AI tools, your competitors will get this knowledge faster. So, you should not relax”.
The CAIO highlights companies with a limited number of employees with revenue in the “hundreds of millions”, questioning “Can some other companies repeat this success? Absolutely. Can they do something similar to SOFTWSWISS? Absolutely. But that’s not only a risk, it’s also a very good opportunity”.
To allow this growth, employees have to be allowed to experiment, but “first of all, you need to understand the constraints, limitations, and capabilities of your tools. Your tools are the LLMs (large language models) and the agentic harness that is built around these models. Then understating “your prompts, your context, and the way that you build your workflows”.
This encompasses “additional guardrails” and checking the work results of the AI tools – with even this checking process being automatable after some time.
But “for some cases, of course, you need the human touch”, and yes, “sometimes AI can hallucinate”, and different models can react differently, or the same prompt can generate different results. But this is mitigated by the ongoing improvements by the AI tool developers themselves, as well as SOFTSWISS employees getting “more and more experience in how to build these pipelines with AI, how to provide better guardrails, prompts and build better context”.
Making sure nobody does anything wrong involves specific guidelines, guardrails and governance, with Romanovskiy noting that “we have a small team, but we have a set of champions who help us: in software development, IT services and business organizations we have these people who work closely with me, with their teams, and help them to get enabled”.
Best use cases
“Is it hype? Absolutely. Is it not hype? Absolutely”.
“The implementation is cheap; something that would have taken a couple of months now takes a couple of hours”.
But where should an iGaming operator be focusing its efforts when using AI?
“We work with a lot of players, and players have multiple different situations where they need help: a lot of communication about something wrong with the games, payments, KYC processes and so on. The cost of support is high. At the same time, the amount of players is growing and the check from a single player is shrinking. We have hundreds of thousands of players that we need to process very quickly, especially in areas like Asia, LatAm, Africa, etc.”
That means that this is the “hottest area”: “how can you build a proper chatbot, a kind of Avatar for technical support that can reliably help the player with their situations? That’s one of the biggest problems”.
Compliance is also key. “In compliance you need to prepare your data properly, formalize it, you need the proper format, to ensure everything is done according to the compliance regulation. And that’s where AI can be helpful. You can automate all these rules, you can check your artifacts, you can ensure everything is in place and save a lot of time for yourself, for your auditor, and for the regulator”.
In addition: game content. “We need a lot of content; we produce a lot of content and it’s very expensive. With AI, we can dramatically make it cheaper. And a lot of companies in game development are already taking AI very seriously in content creation”.
If everybody looks the same, will we get tired of looking at each other?
Does that mean that all the content will start to look the same if creators are all pulling from the same LLMs?
“There will be a lot of content, good and bad. But then, there will also be differentiation. It will be not about the ability to produce content, but if you can build good pipelines to get better, fresh ideas. If you can visualize something, you need good artifacts where you can get ideas from. And if your people understand your audience better […] this will be your differentiator. This and the protection of content”.
“Maybe people will not care if it’s AI generated or not. They will still care about the quality; they will care that the people behind what was created understand them.”
But does this also mean that a company like SOFTSWISS can become entirely autonomous? Romanovskiy has his doubts.
“You can build an autonomous workflow and make some kind of business models that work autonomously for a certain period of time. But the world is changing very quickly, and you need to learn every day and adapt every day. Can the systems do that? At the moment, absolutely not. There are no such capabilities. The existing LLMs cannot train on the fly, they cannot learn from what you’re doing. You have to kind of emulate it with context engineering.”
Even if larger companies are working with better LLMs, “how soon will they put it on the market? Probably decades.”
“Then you need memory. The systems need to learn, so they need to remember, they need to forget, they need to optimize the data. Do they have such capabilities? Not yet. Do we know how to build them? Absolutely not”.
“Then reliability. The systems should be absolutely reliable, they need to work predictably. Can they do that now? No, they can’t, it’s still an area of development.”
While that is something that is likely to see strong improvements this year, complete automation is not just around the corner.
“Do we need companies to be fully autonomous? I don’t think so. We need them to be faster, to be more flexible and probably to make some decisions by themselves. But we still want to rule because we don’t trust this technology enough.”
“You need a person who has a brain, who has motivation, who is part of society and knows what is good for other people and what is not. [AI] is not there yet”.
Denis Romanovskiy, Chief AI Officer at SOFTSWISS
