MarVi: Integrating AI - powered virtual assistant for Marriott Bonvoy
Project Duration
February 2025 - April 2025
15 weeks
Skills
Usability Research
Conversational AI
LLM, NLPs
Tools
VoiceFlow
Figma
Miro
Problem
Currently, for Marriott Bonvoy's website, users experience challenges with the visibility of the booking process, inflexible pop-up screens, and a lack of clear system status or user direction. Moreover, there’s insufficient feedback or reassurance when making changes in the booking portal.
Solution
Introducing MarVi - an AI powered virtual assistant, making booking hotels with Marriott Bonvoy faster, easier, and more intuitive. Marriott can continue to exceed customer expectations, fostering trust and loyalty, and reinforcing its commitment to excellence in hospitality and service.
Virtual Assistant
Meet MarVi - Plan your whole stay with one conversation!
Evaluating the need of an AI chatbot
In today’s fast-paced digital landscape, AI chatbots have become essential for enhancing customer experience, streamlining operations, and improving engagement. For a global hospitality brand like Marriott Bonvoy, integrating an AI chatbot is not just an upgrade—it’s a necessity.
Improving Booking Process
Ensures clear filter visibility and helps users refine choices easily and provides real-time feedback on applied filters and booking status.
Boosting Customer Loyalty
By keeping up with the digital trends, Marriott provides digital assurance and a frictionless travel planning experience.
Building a Competitive Edge
Keeps Marriott on par with or ahead of competitors using AI while meeting modern traveler expectations for convenient, AI-driven service.
Usability Tests to analyze Marriott's Digital Landscape
“Unmoderated Usability Testing : A Deep Dive into Marriott’s Digital Experience”
To ensure the chatbot seamlessly integrates with Marriott’s model, we conducted a thorough analysis of -
1. booking flows
2. search functionality
3. filter usability
4. user pain points
Understanding these elements helped identify gaps and opportunities where the chatbot could enhance efficiency, clarity, and overall user experience.
Unmoderated Interviews Methodology
Process: Where Users Roam Free and Feedback Gets Real
Usability test findings
4 Major Issues identified– And That’s 4 Opportunities to Get Better!
User confusion
Due to unclear visual hierarchy and scattered navigation, users often don’t know where to start or what to click. Whether they're checking points, modifying a reservation, or exploring hotels, the path needs to be made more intuitive, causing unnecessary friction.
Low feedback
After users take an action (e.g., applying a filter, changing currency), the system often provides minimal or delayed visual feedback. This lack of reassurance leads to uncertainty about whether their input was accepted or processed.
Too many functions
From managing bookings to comparing rates, viewing deals, joining Bonvoy, or redeeming points — everything lives on the same screen. While powerful, this leads to cognitive overload and makes the interface feel cluttered instead of purposeful.
Irregular Interfaces
Different pages on the site (e.g., homepage vs. booking engine vs. loyalty dashboard) often follow different design systems. This lack of consistency makes the experience feel disjointed and reduces user trust.
Now, let's Build - A - Bot!
Designing the personality
Why Personality Matters? Personality Powers Connection – Even in a Chatbot!
A chatbot without personality is like a hotel lobby without a concierge — functional, but forgettable. Personality turns your bot from just another support tool into a memorable brand touchpoint. It builds trust, boosts engagement, and makes users feel like they’re talking to someone who actually “gets” them. It’s not just about tone — it’s about experience. Hence, Marvi - Professional and playful - Helping users adapt and keep the fun alive.
Structuring the Brain: Flows, Intents, and Functions
Next, I created flow diagrams and intent mapping to clearly define the chatbot’s functionality, organizing user inputs into structured paths for seamless, goal-oriented interactions. This step plays a crucial role to streamline and simplify the conversation architecture in Voiceflow, efficiency in the development process.
I decided to build two scenarios - 1. Users booking a hotel room for personal use; 2. Users booking the hotels as an event venue. The two scenarios require different conversational approach as they have very different needs and set up.
Scenario 1 - Hotel Room booking process.
Scenario 2 - Hotel Booking process for an event
Setting the structure in Voiceflow
From Flowchart to Full Conversation — One Block at a Time
In Voiceflow, I translated the mapped flows into an interactive prototype by integrating core functions—talk, listen, logic, and developer blocks. This allowed me to simulate real user interactions, refine responses, and ensure the bot delivers a seamless, human-like experience.
Setting up interaction flows in Voiceflow.
Prompts within "Talk".
Prompts within "Listen".
Prompts within "Logic".
Prompts within "Dev".
User Testing and Figuring Anomalies
Testing Conversations with 18 users — One text at a time
Assessing the purpose and function of each Voiceflow block was essential to ensure clarity in interaction design. It helped determine the right point of use, enabling smoother transitions, logical flow, and more meaningful, goal-driven conversations throughout the user journey.
Feedback from usability testing
Key Insights
Transforming Feedback into Forward Motion
Usability feedback revealed recurring themes across conversations, helping us identify key areas for improvement. These insights were grouped into four categories—Feature Opportunities, Intent Understanding, Token Reliability, and Successful Interactions—each highlighting a different facet of the user experience.
1. Opportunities to Integrate Feature
Explore adding image cards, filterable queries, calendars and make interactions more visual, and task-oriented.
2. Language and Intent Classification (WIP):
Improve the chatbot’s understanding of diverse phrasing, regional language to boost accuracy in recognizing user intents.
3. Voiceflow Token Emergencies:
Address token management issues during high-traffic periods or handovers to prevent broken experiences or loss of user progress.
4. Interactions That Work:
Scenarios where the chatbot reliably completed tasks, and satisfied conversational flow.
Final Chatbot
Final Voiceflow output with glimpse of behind the scenes workflow integrated with feedback
Here's MarVi, navigating hotel and event bookings - flex, glitch and fix its features like a true digital concierge."
MarVi: Desktop version
MarVi: Mobile version
MarVi: Final Workflow
Sneak Peak
A sneak peak into our archives
