Game changer
Frodx
Gorazd Zakrajšek
Kinetara:
What It’s Like
Talking to an
AI Agent
While many companies are still figuring out how to use basic AI, FrodX is building its own platform: Kinetara. This isn't just another chatbot. We're talking about advanced voice agents that can handle entire processes on their own – from dealing with missed calls to collecting consent for e-invoices. We sat down with Gorazd Zakrajšek, FrodX's innovation advisor and one half of the RASTezanje podcast duo. We wanted to know how Kinetara works, what the biggest development challenges are, why the future is about improving customer experience (not just cutting costs), and of course – what the hell is “agentic BPO.”
TABLE OF CONTENTS
loading...
Let's start at the beginning. What exactly is Kinetara?
Kinetara is basically our internal AI product built on voice technology. It works as a conversational agent. The approach we use is called agentic BPO – Business Process Outsourcing. What that means is we're not just focused on the technology itself (which is comparable to what's out there), but on having the agent take over and execute an entire business process from start to finish. Kinetara is completely new in this region. And even if you look much further abroad, across an ocean or two, the market is basically split into two groups that, in my opinion, don't address outsourcing challenges holistically. Tech startups usually offer good but really narrowly specialized solutions. Traditional BPO firms rely on human staff and only gradually integrate AI, which makes them less flexible, slower, and usually more expensive. But here's the thing: nobody takes responsibility for how AI agents actually perform. Nobody guarantees results. Kinetara works differently. We take full responsibility for the agent's performance and success. Our team monitors it and steps in when needed. There's always a human in the loop.
Can you explain in more detail how it works? And what's your role in the project?
Kinetara is built modularly, which means each module is a specific agent covering a particular part of the process. Right now our main focus is on the Missed Call Agent. The core problem we're solving is really common: during business hours, companies often can't answer all their calls, which leads to lost opportunities and poor customer experience. When that happens, our agent picks up, has a natural conversation to gather key information – who's calling, why, and when would be a good time to call back – and then sends everything to the company as an email summary and SMS reminder.
My role in the project is mainly on the business and content side. Together with Andrej, who covers the technical part, and Igor, who handles sales, I coordinate the development and implementation of these solutions.
My role in the project is mainly on the business and content side. Together with Andrej, who covers the technical part, and Igor, who handles sales, I coordinate the development and implementation of these solutions.
You mentioned the missed call agent. What other areas can you help companies with?
We also use Kinetara to collect consent for e-invoices under GDPR. To gather meter readings. For taking orders and reservations. This is especially relevant for companies that issue a lot of physical invoices – utilities, property managers, that kind of thing. We start a conversation through SMS, offer users the switch to e-invoices, and along the way handle any missing consents.
We're also developing an agent to help with online shopping, but it's different from typical chatbots. Our goal is to simulate a conversation with a salesperson in a physical store who understands customer needs through dialogue. Like, is someone a mountain biker who needs top-tier gear, or a casual rider looking for a hybrid bike? We can then use that information for personalization in other systems.
We're also building an After Offline Meetings agent that calls a field agent after a meeting and walks them through key questions, then automatically logs the answers in the CRM. Think about a real estate or insurance agent who's out in the field and just doesn't have time to take notes during conversations.
We're also building an After Offline Meetings agent that calls a field agent after a meeting and walks them through key questions, then automatically logs the answers in the CRM. Think about a real estate or insurance agent who's out in the field and just doesn't have time to take notes during conversations.
There are tons of AI solutions on the market. What's Kinetara's main value add? Are you more focused on cutting costs or improving customer experience?
Our primary focus is improving customer experience. For small and medium-sized businesses, that's what
matters most. Let me give you an example. I don't know a single person who likes those annoying automated answering systems where you're pressing numbers and getting bounced around between different agents. Kinetara is going to eliminate that night mare.
matters most. Let me give you an example. I don't know a single person who likes those annoying automated answering systems where you're pressing numbers and getting bounced around between different agents. Kinetara is going to eliminate that night mare.
"Nobody takes responsibility for how AI agents actually perform. Nobody guarantees results. Kinetara works differently. We take full responsibility for the agent's performance and success.
Instead of users pressing numbers to select options, the bot will simply ask: “How can I help you?” and based on the answer, route them to the right person. The caller's experience improves dramatically.
So costs aren't that important?
Of course, in large systems like call centers, the cost reduction angle matters too. Our agents can handle a good chunk of inbound calls, which lets human agents focus on more complex outbound calls. This shows up in our model, which is – brace yourself for a bunch of acronyms – KPI-backed, SLA-owned, pay-for-performance. That means we charge per minute used as a baseline, but we only charge extra when a call leads to a concrete action: a reservation, obtained consent. If our agent successfully books an appointment at a dentist or auto shop, we charge an additional amount for that successful action. Same with e-invoices; for every customer we convince to switch to e-invoices or where we obtain GDPR consent, we charge extra. Our success is directly tied to the client's success.
Developing an AI agent that speaks multiple languages probably isn't easy. What challenges did you run into during development?
True, it's not just about Slovenian. Our goal is to support eight major languages in the Central and Eastern Europe region. Right now the experience of switching between languages is pretty entertaining. If you start a conversation with the agent in Slovenian and ask it to switch to Croatian, it'll speak Croatian with a Slovenian accent. And vice versa.
But yeah, there are a ton of challenges. One of the main ones is that the dataset for smaller languages in LLM models is incomparably smaller than for English. That affects stability. If we give the module too much creativity in its responses, interesting anomalies can pop up – it starts generating weird letters or drops an English word into the middle of a Slovenian sentence. But if we limit creativity too much, speech gets slower, which bothers some people.
"People are surprised by how natural conversation with AI has become. The fear that you wouldn't be able to have a real conversation with a bot is disappearing.
The ability to process transcripts is also really interesting. Sometimes they look like a complete mess of words, but AI still manages to create a correct and sensible summary from them. It handles different dialects surprisingly well too, which is crucial for wide usability. That's part of the magic behind the scenes that even we don't fully understand.
We've also found that it really calms customers down when the bot confirms it understood them correctly. We pay a lot of attention to this, so the bot will sometimes repeat the caller's last words to make sure there's no question about whether the message was understood correctly.
Finally: what motivates you most about working on this project?
AI has become part of my daily life, both at work and in my free time. I feel like this is a field where you have to constantly play around and explore if you want to stay current. I use AI for everything, from my kids' homework assignments (which are getting more complex), to composing music. Even with our podcast with Andrej, part of the process – like distribution and content slicing – is already completely automated with AI. That constant play and discovery of new possibilities is what I find most interesting.
Številka 4 | DECEMBER 2025
GAME CHANGER
INSPIRING BUSINESS STORIES OF REMARKABLE PEOPLE
