David, in the energy sector, which maybe gets unfairly labeled as pretty conservative, you were among the first to seriously adopt artificial intelligence. When did you realize AI was non-negotiable?
GEN-I isn't just an energy supplier – we're also a trading and analytics firm. Trading requires massive amounts of analytics. That's why we've always collected and analyzed insane quantities of data. And that's why we were using AI long before the LLM boom. We analyze sentiment on social media, read news feeds, use satellite images to determine water levels in hydroelectric plants. We've been working with traditional machine learning since the company's early days. Bottom line: we didn't just discover AI when OpenAI showed up
So we're talking about two sides of AI: numerical analytics for trading, and in recent years, generative AI for customer communication. How did you tackle that second area?
Exactly, these are different things. In trading, AI helps us gather data on partners, summarize business reports, recognize visual materials, and flag special risk factors. This improves the accuracy of our market movement forecasts, consumption, production – which is essential for GEN-I's competitive edge. In plain English: it lets us be cheaper. When it comes to LLMs, we see opportunities mainly in customer support, sales, and back-office functions. That's why we first tackled email triage with FrodX, where we immediately hit 85% accuracy. But it's not about AI directly responding to customers – it's about sorting and routing emails, and in certain cases helping prepare a draft response that an agent then completes and sends.
Next comes automation where possible. If a customer says they lost their bill, we don't need AI to send them another one. The same automated process works for both an AI agent and a customer support employee, solving the customer's problem in seconds. We really need artificial intelligence for more complex questions. Say a business client is missing their meter number for a quote. The AI prepares a draft response asking for that info, which the agent then sends. The agent still reviews and completes every response. But the end goal of the pilot project is for AI to understand the message intent and prepare a quote draft for the salesperson.
Next comes automation where possible. If a customer says they lost their bill, we don't need AI to send them another one. The same automated process works for both an AI agent and a customer support employee, solving the customer's problem in seconds. We really need artificial intelligence for more complex questions. Say a business client is missing their meter number for a quote. The AI prepares a draft response asking for that info, which the agent then sends. The agent still reviews and completes every response. But the end goal of the pilot project is for AI to understand the message intent and prepare a quote draft for the salesperson.
"If we buy too little energy for someone, we'll have to buy more at the last minute at a higher price. If we buy too much, we've got a problem again. A good deal for us is a good deal for the customer.
You mentioned business clients. Interesting that you focused on them, even though there are way more residential customers. Wouldn't it make more sense to implement AI for quote preparation with the latter?
The answer's simple. Residential customers get two or three price plans. It's mass sales where ease of use matters most – usually just through a portal from your couch. With business clients, the story's way more complicated. The biggest ones obviously have their account managers who take them to meetings, lunches, coffees, and eventually agree on a fair price. Then we've got several thousand small and medium businesses. We can't have account managers for them because there are too many. But they definitely deserve better treatment than just two generic price plans. That's where AI is key. With its help, our account manager can handle way more clients with quality service. In energy, a personalized offer means a better deal for both sides. If we buy too little energy for someone, we'll have to buy more at the last minute at a higher price. If we buy too much, we've got a problem again. A good deal for us is a good deal for the customer.
You mentioned you have no plans to reduce the number of call center agents. That goes against the general belief that the goal of AI automation is cutting labor costs.
That's right, and maybe that's the most important message. We have no ambition to lay off agents. We have a call center in Krško where people are happy to work for us. We don't have problems with turnover or onboarding new employees. We're excellent at this. But the main thing is that a good specialist who grows in the call center and understands customers is excellent talent for other departments – in sales, marketing, tech support, or IT, where they become great business analysts. This philosophy shows in our HR policy: hiring additional agents isn't hard because we have solid onboarding and knowledge transfer processes. We do this to speed up processes and create more time for real consulting, quick and expert solutions, and more personal treatment. With digitalization and automation of standard processes, our people get more time for consulting and complex tasks.
"Digitalization doesn't mean we'll do the same thing digitally. It means we automate standard processes so we have more time for what's not standard.
How do the AI technologies and philosophy you described affect the future? What are your next steps in product development?
Our ambition is obviously expansion into foreign markets and development of modern energy products. At the core of all these products is flexibility. This means the customer adjusts their consumption and production not just to themselves, but to the entire market. If you can use more energy when there's lots of sun, you've made a good deal. It's increasingly common for energy prices to go negative. That means you get paid to use it. We can barely imagine modern energy supply without connected devices – from batteries to cars – that can recognize your habits, weather, and market scenarios and adjust consumption and production for greater self-sufficiency and more efficient grid support. Backend systems also need to be efficient and flexible, as new products and payment methods are emerging that gravitate toward subscription models. You basically subscribe to electricity, just like you increasingly do with other services. We've been doing this since before Netflix.
Why did you choose FrodX for this project, given that there are tons of tech companies out there?
Let me give you an example – I'm basically an IT guy, but I've done lots of things. Among other things, I was marketing manager at Microsoft, where I learned a ton about sales, working with customers, support. I think that's similar to laying tiles. It's not hard to lay tiles – what's hard is making nice grout lines. The seams that connect different areas and departments. Same goes for FrodX. Even though using AI is also a technical problem, it's really more of a human problem. We've got endless offers from tech firms. But FrodX is one of those companies that understands the customer, user experience, business, marketing, and branding. You can't use
technology like AI without understanding these things. Otherwise a tech company tells you: “You'll fire everyone in the call center.” But that's not our goal. A bot will never be better than an agent. Nobody, ever, will call FrodX or GEN-I and ask if they can talk to a bot.
technology like AI without understanding these things. Otherwise a tech company tells you: “You'll fire everyone in the call center.” But that's not our goal. A bot will never be better than an agent. Nobody, ever, will call FrodX or GEN-I and ask if they can talk to a bot.
Številka 4 | DECEMBER 2025
GAME CHANGER
INSPIRING BUSINESS STORIES OF REMARKABLE PEOPLE
