Scout24, Germany's largest real estate platform, decided that search alone wasn't enough. Looking for an apartment is just the beginning; people want guidance, advice, and clear answers as they sift through listings, agents, and options. Sound familiar? That was the spark behind their new conversational assistant powered by GPT-5.
How the new assistant works
The Scout24 team built HeyImmo, a conversational assistant that doesn't just return listings: it asks to clarify your needs, shows results with images, and adapts how it responds based on your preferences — short summaries, key bullet points, or direct lists with previews.
It's not magic, it's experience-centered design: if you want bullets because you're comparing neighborhoods, you get bullets. If you prefer a narrated summary of the top options, the assistant provides that. Want to see apartments right away? It shows those too.
Simple technical decisions to learn fast
Although they initially explored complex architectures with multiple agents, they ended up betting on simplicity: function calling and focused components that let them iterate and ship fast versions.
Technical priorities were clear: speed, relevance, and well-structured answers. If the app responds slowly or confuses you, you get lost. That's why optimizing latency and consistency was essential.
How they measured quality
A central part of the project was defining what a good answer actually means. For that, they created their own evaluation system inspired by OpenAI Evals. They didn't settle for intuition; they defined metrics and measured performance in real scenarios.
They also ran a massive internal test: almost the entire company tried the product, reported bugs, and pushed edge cases. The result? The company even delayed the launch to meet the quality threshold they had set.
Learn by doing, but with guardrails: iterate fast and know when to pause.
Collaboration and safety
Scout24 worked hand in hand with OpenAI solution architects to raise answer quality and align the experience with safety and trust criteria. That collaboration sped up decisions about response structure and risk mitigation.
It's a good reminder: you don't have to do everything alone; technical partnerships can save months of trial and error.
Practical lessons you can apply
- Start with what transforms your core product. For Scout24 it was search.
- Define what quality means for your case and build tools to measure it.
- Invite the whole team to try early and often.
- Iterate quickly, but stop if the experience doesn't meet the standard.
These are simple guidelines, but hard to keep without discipline. Scout24 applied them and launched an experience that aims to be useful and reliable.
Where they're headed next
Scout24's focus isn't only on seekers. They want to expand HeyImmo for owners, landlords, and agents: from verified floor plans to personalized guides based on user type. The idea is to deepen the experience and leverage network effects across their ecosystem.
As the CTO says, they're just getting started: many AI-driven features are on the roadmap.
Final thought
This story shows something many companies still need to learn: the most powerful AI doesn't replace a good product — it amplifies it when integrated with rigor, testing, and human judgment. Want to improve your search? Start by listening to your users, measuring quality, and building an experience that actually helps them.