Challenge
Bezrealitky wanted to make their customer support more efficient and improve the success rate of query handling. The main goal was to prepare for managing a significantly higher volume of communication in connection with their planned expansion—without having to increase existing team capacity.
- The original button-based chatbot, powered by a simple NLP model, could automatically resolve only 65% of the most common queries.
- A major challenge was the need for frequent manual retraining of the model on new data.
- The objective was to develop an intelligent AI chatbot capable of handling most queries without complex maintenance, with the ability to automatically update its knowledge from various sources (website, blogs, call transcripts, static documents, etc.).
Our Solution
We deployed an advanced AI chatbot and a follow-up AI voicebot, both leveraging internal knowledge sources and the GPT-4o language model to deliver accurate, context-aware responses.
The success rate of the AI chatbot’s automated responses increased from the original 65% to 93% within the first three months after launch. With the addition of the AI voicebot, automation was extended to phone calls as well – with only 9.1% of calls requiring operator intervention. Thanks to the Feedyou platform, Bezrealitky can now use a unified knowledge base and conversational logic across all channels, enabling faster implementation, greater consistency, and lower costs.
,,The AI chatbot on our website, Bezrealitky.cz, can handle the vast majority of common user inquiries without involving a human operator.
This has effectively saved operators’ time, allowing them to focus as specialists on cases where a human touch and relationship are simply indispensable.”
Martin Ponzer, CEO @ Bezrealitky
Hear the AI voicebot in action
Thanks to advanced NLP, multilingual support, emotion recognition, and integration with your systems, the voicebot can hold natural conversations, respond to context and the caller’s mood, and intelligently transfer the call to a live agent when needed.
Step-by-step implementation
Current state analysis and goal definition
We started by analyzing the existing chatbot, which relied on a button-based interface and a limited NLP model.
We quickly identified key issues: understanding only 65% of queries, high manual effort to train the model, and no automated updates to its knowledge base.
Together with Bezrealitky, we defined our goal: an intelligent AI chatbot that can handle most questions on its own, learns automatically from data sources like the website, and drastically reduces the need for manual intervention.
Integration of an advanced AI model
We replaced the original solution with a more robust language model capable of handling open, natural conversations—without relying on rigid buttons. Quick replies and buttons are used only where it makes sense, such as promoting new features or services.
The chatbot is connected to various internal knowledge sources—websites, databases, blogs, call transcripts, and documents in multiple formats. We designed the conversational logic to feel as natural as possible while covering a wide range of scenarios.
We also implemented a system for regular knowledge updates, ensuring the chatbot stays current without the need for manual retraining.
Expansion to voicebot
Building on the success of the AI chatbot, we introduced an AI voicebot to replace the traditional IVR system (tone-based call routing).
The AI voicebot can hold fully natural conversations and handle repetitive calls end-to-end, such as inquiries about listing status, payments, or the sales process for a property.
We implemented Deepgram technology for faster and more accurate speech-to-text (STT), significantly improving comprehension and response quality.
Moving beyond simple synthetic voices, we now use neural text-to-speech (TTS) with WaveNet, delivering emotionally rich and stylistically versatile voices that can clone a voice or change speech style without complex setup.
Scaling to a multi-agent system
For more complex inquiries, we realized that a single universal chatbot or voicebot with knowledge of everything wasn’t enough. That’s why we developed a multi-agent system consisting of smaller AI virtual assistants, each specialized by domain and expertise.
Different AI assistants focus on specific areas, such as technical support, billing, or legal questions, and can hand off inquiries based on topic or context.
Handoffs can also occur by language—for example, a Czech-speaking voicebot can transfer the conversation to an English-speaking bot with a different voice and knowledge base, ensuring the interaction feels natural.
This approach allows us to cover a wider range of queries that can’t easily be categorized into a few common topics. By having each bot work with a smaller, specialized dataset, we reduce overlap, hallucinations, and latency. The result is faster and more accurate responses, lower costs, and simpler system management.
Key metrics from the AI chatbot pilot
Most inquiries are resolved on the first try. The chatbot handles thousands of conversations monthly, learning continuously from web content, transcripts, and documentation—allowing the support team to manage more without extra resources.
Key metrics from the AI voicebot pilot
,,Working with Bezrealitky has been an exciting journey. They needed an AI solution that could handle the majority of customer interactions without constant manual retraining. We delivered an advanced AI chatbot and AI voicebot system that learns from multiple data sources and understands natural language, boosting successful query handling. This partnership highlights how intelligent AI solutions can empower teams.”
Vojtěch Dlouhy, CEO @ Bezrealitky
Advantage of using the Feedyou Platform
Bezrealitky use the Feedyou Platform as a unified solution for chat, voice, and email communication. This allows them to reuse the same knowledge base, scenarios, and conversational logic across all channels.
When customers interact through multiple channels (e.g., chat and voice) within a single platform, it creates better synergy – knowledge and technical components are reused, enabling faster implementation, higher success rates, and a consistent user experience.
The Feedyou Platform thus serves as a comprehensive tool for intelligent customer service automation across digital channels – eliminating unnecessary duplication, information loss, and extra costs.
About Bezrealitky
Bezrealitky have been operating since 2007 as a leading online platform connecting property sellers and buyers directly. Originally a Czech project, the company has gradually expanded to Slovakia, Germany, and Austria.
Despite a compact team of around 20 people (including 3 in customer support), they handle approximately 2,500 emails, 3,000 chat conversations, and 1,000 calls per month. Thanks to a focus on automation and smart digital processes, they efficiently manage high volumes of inquiries without increasing staff.
With an annual turnover of around 200 million CZK, the company continuously invests in technology and development. Leveraging a unified platform and strong partnership within the EHS group, Bezrealitky is recognized as an innovator in the digital transformation of the real estate market in Central Europe.
More inspiration
A voice assistant that helps you automate communication where the keyboard alone isn’t enough.
