Detour me, Pierre!

 

Long story short

 

Context: Pierre’s goals (speed, efficiency) and the user’s goals (assurance, curation) have been misaligned since Brain B&b's organisational restructuring. This misalignment has left both new and returning users disappointed and is negatively affecting how users engage with Pierre. Pierre was previously known for its ability to thoughtfully curate stays. The priority now is to win back the eroding trust of returning users in Pierre’s curation capabilities and to attract new users to Brainbnb

 

Underlying anxiety: Studies on AI travel tools and chatbots find they are mainly used for inspiration, rough planning, and getting a sense of options, while humans, and travel websites remain the preferred channels for final booking and payment (Orden-Mejía et al., 2025). Among the smaller group who book through AI assistants like Pierre, users report higher satisfaction with human agents than with a competent bot that feels cold and lacking in warmth (Zhang et al., 2025).This means, competence alone is not enough; if Pierre’s language and User Interface signal efficiency but not care, users will feel underserved and either switch channels or avoid using Pierre next time, even if the suggestions are technically good

 

Rationale: Brainbnb users drop off at the listing page not because the stays are wrong, but because the page gives them no signal that the listings are right. Information scent describes the cues that help users predict whether continuing will get them closer to what they want (Nielsen Norman Group, 2020). This concept builds on information foraging theory, which shows that users navigate digital environments like foragers tracking prey, following paths that show the strongest signal of reward and abandoning those that do not (Pirolli & Card, 1999)

 

When Pierre shows results without reflecting the user's own words back to them, the scent collapses. Users read this as Pierre not listening, and they leave. This study strengthens information scent end‑to‑end: from how Pierre frames the conversation, to how recommendations are explained, to how each page surfaces relevance, context and social proof at a glance

​

What I did: 

​

Systematic Literature Review On Chatbot & Machine Anthropology → Logistic Regression On Booking And Chat Logs → Sentiment And Topic Analysis On Conversations → Thematic Clustering And Phase‑wise Friction Mapping → B = Map Framing → Competitive Scan And Red‑route Prioritisation 

 

Duration: 5 days 

 

Tools used: Google Colab | Claude | VS Code | Adobe XD | Canva | Figma

​

View full report and video of wireframes with microinteractions here