The AI travel assistant launched end of 2024.

Summer of 2025, Tripadvisor did a brand-refresh. So all original colors and components were recently (or in the process of being) updated.

The AI travel assistant is an AI chat tool that can be accessed across the site. The tool can help a user get ideas for an up coming trip, ask questions, and save it to your upcoming trip.

This initiative was to work with the team to identify core user and business problems to solve for.

1. Logged-Out User Engagement Gap
Most users interacting with the FAB are not logged in, which leads to lower engagement and a steep drop-off in chat completion rates.
   
   2. Unclear FAB Value Proposition Inside Chat
Once users open the chat, Ollie’s purpose and benefits are not clearly communicated or reinforced, causing many users to lose interest before taking action.
   
   3. Weak In-Chat Dynamic Prompts
The prompts shown inside chat are less relevant and engaging compared to in-page prompts, resulting in fewer users starting or continuing conversations.
   
   4. Device-Related Drop-Offs
User drop-off rates are higher on certain devices—especially mobile—likely due to chat layout or usability challenges that disrupt the browsing experience.
   
   5. Performance Varies by Page Type 
Engagement and completion rates fluctuate significantly depending on the type of page or entry point, suggesting inconsistent user value or experience across different site areas.

In-chat dynamic prompts are less effective than in-page prompts 
(Supported by data)

• On destination pages, 70% of queries are initiated via prompts, while in FAB only 40% are.

• Visibility and context appear to be stronger for in-page prompts, suggesting in-chat prompts may be less compelling or relevant.

High proportion of logged-out users is limiting engagement 
(Supported by data)

• The FAB rollout is occurring in an environment where 85% of interacting users are logged out, versus 69% in the pre-test on destination pages.

• Since logged-in users typically have higher engagement and completion rates, this shift in audience composition likely contributes to the lower overall conversion from FAB click to message sent.

Value proposition of FAB is not clear enough 
(Partially proven)

• The CTR is low (0.24%) and once users open the chat, there’s a big drop off in messages sent (7.4%), indicating that many users do not see enough reason to engage or understand the FABs purpose. 

• The number of users who click on the sticky header (collapsed entry point) is twice the number of users who click on the FAB on destination pages. This suggests users may not be perceiving the value of the FAB.

Initial curiosity is driving early engagement, but not sustained use 
(Likely, but needs further testing)

• 51% of FAB interactions happen on first exposure, coinciding with the tooltip display.

• This suggests the tooltip prompts curiosity-driven clicks, but without continued visibility or additional triggers, repeat interactions are limited.

• To confirm this, we could test scenarios such as removing the tooltip, varying its design, or reintroducing attention triggers (e.g., periodic animations, reminders) and observing whether engagement beyond the first impression improves.

Servlet-level differences may be impacting results 
(Partially supported, needs deeper analysis)

• Some servlets are not showing proportional growth in query initiations and message sends relative to impressions and clicks.

• This may relate to differences in page intent, layout, or audience composition.

Device differences may also be impacting the results 
(Partially supported, needs deeper analysis and more volume of data)

• Mobile web engagement is consistently lower than Desktop/Tablet web across all steps of the FAB funnel, except for FAB click. 

• This could have to do with the fact that the chat window takes up the full screen on Mobile web, unlike Desktop where travelers can continue to browse while they chat.