| OVERVIEW
At Trivago we noticed a clear gap in how users search: they often choose places based on how they look, yet finding the right vibe means scrolling through endless listings. Style matters, but it’s hidden.
The product didn’t offer structured data around hotel aesthetics. This opened up an opportunity to explore how AI could surface style in a scalable way and make visual discovery feel effortless.
What we noticed:
Users want a faster, easier way to find hotels that match their style preferences.
| RESEARCH INSIGHTS
Image-Based Decisions
People choose places based on how they look. Visuals drive decision-making, and users want a way to filter options by style such as cozy, minimalist, or luxury.
Time-Consuming Search
The current process of browsing through a long list of hotels to find one that matches their style preference is tedious and inefficient.
Lack of Style Information
The product lacked structured data around hotel aesthetics, which made it hard to surface style as a filter or help users choose with confidence.
| OPPORTUNITY
By utilizing AI to analyze hotel images and categorize them by style, we can offer users personalized style-based recommendations.
People choose places based on how they look. Visuals drive decision-making, and users want a way to filter options by style such as cozy, minimalist, or luxury.
| PROPOSED SOLUTION
Create an AI-powered module that recommends hotels based on user-preferred styles. The module appears mid-search results, includes style-based tabs, and lets users compare hotels with a similar vibe.
| PROCESS
Building AI-powered style recommendations into the backend would have required a significant engineering investment. Before committing to that, we needed to test a key hypothesis:
Would users engage with hotel suggestions based on visual style?
To answer this, we launched a live Proof of Concept.
We implemented a real experience directly into Trivago’s search results, using hardcoded hotel data to simulate backend functionality.
01
Research
Partnered with UX Research to identify which hotel styles resonated most with users and evaluate which images best matched users’ style expectations. This helped us define a shortlist of high-signal aesthetics.
02
Development
Worked with frontend engineers to build a real, styled module. We manually selected 3 hotels per style, reused and lightly adapted existing components, and integrated the module into the actual search flow using hardcoded data.
03
Testing
Launched an A/B test to compare user engagement and evaluate behavioral impact, including refinement and conversion metrics.
04
Analysis
Collected real user insights with minimal backend effort, validating the concept’s value and guiding future investment decisions.
| OUTCOME OF PROOF OF CONCEPT
High engagement with the module and style tabs
Clear user interest in visual recommendations
Signal strong enough to justify backend investment
| NEXT STEPS & IMPACT
After the success of the module, scaling image-based recommendations was still a big challenge, but we found another way forward. The backend content team used AI to tag hotels by style based on guest reviews, unlocking a second A/B test.
This time, we added style filters to the main filter menu.
Users responded well again, the filters test was accepted and is now part of the core product.
| KEY LEARNINGS
Style is a decision driver
Visual appeal isn’t just aesthetic. Users rely on it to assess relevance fast, it’s how they decide.
Modularity is powerful
Reusing components let us test fast and scale ideas quickly into the core product.
Low-effort prototyping can unlock investment
A simple test with hardcoded data gave us enough signal to justify backend development.
