Preserve trust
AI entered as a beta layer, not as a replacement for the established search system.
Back to work
Case Study / AI Search
AI-based search for a mature marketplace.
I designed an optional AI-search beta for ImmoScout24 that protected familiar search behavior while giving product and engineering teams a practical way to learn.
Context
High-intent real estate search with established user habits.
Role
UX strategy, interaction design, testing and product alignment.
Decision
Keep AI optional, reversible and visibly connected to classic filters.
Impact
Valuable AI learnings without negative impact on core search KPIs.
The Tension
Strategic, but unproven.
ImmoScout24 explored how AI could support discovery inside its core search journey. The product challenge was not simply to introduce AI, but to do it in a way that respected a search model millions of users already understood and trusted.
The risk was classic innovation theater: adding a novel interaction without improving the actual decision journey. My design responsibility was to create a testable experience that made room for learning without forcing users to abandon clarity and control.
Design Judgment
The senior decision was restraint.
Keep control visible
User prompts were translated into familiar filters so people could understand and adjust the result.
Learn without pressure
The experience was optional and reversible, which made experimentation safer for users and the business.
Approach
Make AI fit classic search.
The first version treated AI as an alternative input method. Text and voice prompts were interpreted into the existing filter logic, which allowed the team to move quickly while preserving the underlying search behavior.
In usability testing, users found AI interesting but still preferred explicit filters for serious real estate decisions. That insight shaped the product direction: AI should support user intent, not hide the rules of the system.
Research Signals
Users valued predictability more than novelty.
Most users preferred the search model they already knew, especially for high-stakes housing decisions.
It was perceived as technically interesting, but not useful enough for the context.
Users wanted to see how AI interpreted their intent and needed easy ways to correct it.
Outcome
A practical learning platform for responsible AI UX.
Product impact
AI adoption remained limited compared to classical search, but the beta did not weaken the performance or trust of the core journey.
Team impact
The initiative gave product, design, data and engineering teams concrete experience with generative AI systems in a real marketplace context.
Design lesson
AI works best in complex decision products when it increases clarity and control instead of asking users to surrender both.
Next
Explore another product story.
Contact
LinkedIn
