We ran our workshop at the AI Talks by HVG & Amazing AI conference, at the Bálna in Budapest. We had been preparing for those 90 minutes for months.
The room was full, and a good number of people had come to the conference specifically for this session.
What the workshop was actually about
How to build voice-based customer service on AI that does not just work but pays for itself — while the customer experience improves, the operators' days get easier, and the whole operation becomes easier to see into.
We worked through three big questions:
1. When is voice AI worth using — and when is it not?
We showed that AI is not a magic wand for everything. It works well where there is volume, repetition, and a structured process. Customer service automation is a typical case: the effects show up quickly and you can measure them.
2. What does a human operator minute really cost — and what does an AI minute cost?
In the call center world, out of 60 paid minutes only about 25 are actual time on the line. The rest is breaks, wrap-up, admin, and waiting. So the effective cost of an operator minute is considerably higher than it looks at first glance. An AI agent, by contrast, is available around the clock, with no breaks and no queue.
3. What does a 30-60-90 day rollout look like?
We did not only talk about technology, but about the business reality:
- Days 0-30: Pilot — one or two use cases in a controlled setting, and the measurement framework put in place
- Days 30-60: Tuning — iterating on prompts, escalation rules, and KPIs based on feedback
- Days 60-90: Scaling — further use cases, integration into existing systems, full production
The main point: replacing people is not the goal
The central idea of the workshop was that voice AI is not there to replace human operators but to work alongside them. The AI takes the repetitive, low-complexity work; the human colleague steps in where empathy, creativity, or judgment is needed.
In an enterprise setting that matters even more — our enterprise solutions are built on exactly this model.
Live demo: STT + LLM + TTS in real time
We showed how speech-to-text, the language model, and text-to-speech work together in real time. Voice AI that responds in 200-400 ms is no longer IVR 2.0 — it is a different category of communication.
We demonstrated it with Hungarian-language examples in three areas:
- Customer service — automatic questions and answers, opening tickets, escalation
- Lead generation — qualification, booking, data capture
- Level 1 support — fault diagnosis, status lookups, routing
Live poll: where does it hurt most?
During the workshop we asked the room where they feel the most pain. The answers were clear:
- Appointment booking — managing the calendar by hand, every day
- Basic FAQs — repeat questions draining operator time
- Lead generation — qualification is slow and takes people
- Ticket status — customers keep asking, and operators keep saying the same thing
These are the areas where voice AI delivers value fastest. Our use case overview goes through the industry applications in detail.
Where next?
Customer service is heading toward a hybrid model — AI and people, each doing what they are better at. Voice AI is not hype; it is a down-to-earth business tool. There is plenty of room for production-ready voice AI projects, not just pilots.
If you want to know how it would work in your organization, book a call and we will show you the concrete options.