What is Agent AI?
Agent AI is an AI and machine learning (ML) chatbot designed to 1) assist travellers with booking and managing travel arrangements, 2) resolve booking-related issues, and 3) direct users to the appropriate support contact when necessary.
Initially designed to reduce reliance on human support agents and optimize support costs, this product has evolved into a strategic vision for Navan. We aspire not only to blur the line between human and AI agents but also to create an AI-empowered self-serve infrastructure that will revolutionize the travel booking and management experience in the future.
As a staff designer, I work as a strategic partner to C-level managers…
… while overseeing the design work and ensuring the quality of the product.
In addition to daily collaborations with my Scrum team and cross-functional stakeholders, I work closely with the CTO and VP of Design to align visions, track our work progress, and ensure the achievement of our goals and objectives.
Design Owner
I work as the main designer and the experience decision-maker behind Agent AI. I build and scale the product’s Design System. I get and allocate resources and support to move the product’s design progress forward.
Product Visionary
I contribute ideas to develop the product in alignment with its vision, taking ownership of each initiative to validate and propel it forward.
Design Advocate
I facilitate cross-team design and product brainstorming sessions to align Agent AI with the rest of the Navan app, seeking opportunities to leverage Agent AI’s impact across the entire platform.
The challenges
1
We are building something new with limited insights and resources. We must invent new patterns and concepts, testing and iterating until we achieve our product vision.
2
Some users are skeptical about AI due to past negative experiences. Others prefer to speak with a human agent immediately.
3
Users who seek support are generally unhappy and have an urgent problem that needs to be resolved.
4
This product was originally built and pioneered by a team of engineers. Its design was inconsistent, patched with new add-ons and outdated legacy components.
5
Travel support is a vast domain spanning flights, hotels, trains, and other types of transportation. Travel-related issues depend on the type of travel, customer needs, and their stage in the journey.
Main User Personas
Research
I use a variety of quantitative and qualitative research methods to stay data-driven and in tune with our users, collaborating closely with the team’s UX Researcher and Product Owner.
Daily data monitoring
Our team spends 15 minutes in our daily catch-up reviewing the product’s performance highlights. If there is an unexpected increase or decrease in metrics, we discuss it to develop hypotheses and decide on an investigation approach.
Tools: Tableau, Twilio, Amplitude, Snowflake
Weekly analysis
We have a Weekly Analysis session where the entire team comes together to review real chats with Agent AI based on topics, CSATs, research questions, etc.
Tools: Twilio, Sheet
In-depth user research
When a hypothesis, problem, or design solution requires validation, I collaborate with the UX researcher and product owner to conduct comprehensive user research, from planning to analysis. For interview-based research, I alternate with the UX researcher to observe and speak directly with the interviewees.
Tools: Doveteail, Zoom
Usability testing
Sometimes, I quickly test design solutions to gain more insights before finalizing them or conduct A/B tests on different design variants to validate a hypothesis.
Tools: UserTesting, Amplitude
Highlighted projects
Revolutionized support for unexpected airline schedule changes
Problem
An “airline schedule change” refers to a flight booking rescheduled unexpectedly by the airline, beyond the traveler’s control. This situation often causes dissatisfaction, prompting travelers to seek support. Usually, a traveler must navigate multiple channels or contact the airline’s hotline to change their newly assigned flight. This requires time and causes anxiety.
Solution
Agent AI assists by informing the user about the schedule change, explaining their options based on the airline’s policy, and helping them easily reschedule their flight to a more suitable one, all within a seamless chat.
Success
We have increased the rate of successful automated support cases, reduced reliance on human agents (measured by time), and improved both Agent AI CSAT and overall support CSAT.
The most common support request: flight changes
Problem
Travelers sometimes need to change their flights. The conditions for changing a flight vary based on airline policies, cabin classes, the timing of the change, booking status, and other factors. Changing flights can be a hassle, and one of the biggest user needs during this process is cost estimation.
Solution
Agent AI assists by informing the user about their flight booking status and exchange policy, and suggesting the best solutions available. Agent AI also finds the best alternative flights based on the user’s preferences and offers options with the best cost optimization.
Success
Flight exchange is the most common support request we receive. By automating and improving this process, we have achieved significant results. We have increased the rate of successful automated support cases, reduced reliance on human agents (measured by time), improved both Agent AI CSAT and overall support CSAT, and increased flight sales.
Old chatview
New chat experience
Reinvented the chat and inbox experience
Problem
In the previous chat experience with Agent AI, users were transferred between different support agents. This caused anxiety and required them to reexplain their support requests, making the chat experience less seamless. Additionally, Agent AI had an immature brand identity, leading to distrust among some skeptical users.
Solution
I have reinvented the entire chat experience by:
- Establishing a more mature and professional identity that aligns with Agent AI’s capabilities,
- Implementing a seamless and unified chat thread where different agents and specialists join together to support the user,
- Introducing a smart inbox for users to easily track their past and current support requests.
Success
We have reduced the time users spend per support request, increased both Agent AI CSAT and overall support CSAT, and enhanced user trust.
Agent AI’s Design System 2.0
Problem
The original design of Agent AI was developed by non-design engineers with little consistency in its design system compared to the rest of the Navan app. Its UI infrastructure was outdated and not synchronized with Navan’s storybook. Its in UI components were a patchwork of spontaneous engineer inventions.
Solution
I’ve developed a completely new design system for Agent AI, featuring:
- A cohesive Navan design language
- An updated, publicly accessible storybook for cross-functional stakeholders
- Implementation on a new UI infrastructure closely integrated with Navan’s design and other product storybooks
- A more mature and professional design and content language.
Success
Our primary goal with Agent AI is to blur the distinction between human support. We envision a future where users don’t need to be concerned about who assists them, as long as their requests are handled successfully. Since implementing the new Design System and identity, we’ve observed an increasing number of users who interact with Agent AI as if they were speaking with a human, often without realizing or caring whether it’s a bot.
Because we can: Chat Auto-translation
Problem
In addition to our commitment to achieving business goals and objectives, we allocate 20% of our time to inventing fun and innovative experiences for users. This initiative aims to refresh and enhance their overall mental well-being through support—simply because we can!
Solution
One of our ‘Because we can’ projects is Chat Auto-translation, where we enable Agent AI to automatically detect and seamlessly switch between languages based on the user’s chat input, mimicking natural human interaction.
Success
We believe that enhancing user satisfaction through attention to detail has a long-term impact that cannot be measured quickly. It’s about shifting behavior, building trust, and increasing CSAT. We have observed a positive trend in our CSAT scores, with users interacting with Agent AI more patiently and trustingly.
Impact and Success
Significant Support Cost Savings
We’ve reduced our support costs per minute by approximately 40%.
CSAT Improvement
Before I joined the team, the average CSAT was 63%. Currently, our CSAT has increased to around 75%.
Setting vision for the future
This product is reshaping the future of the travel experience, where travel booking and management will be less UI-heavy, more personalized, and time-efficient with the assistance of AI.
Cross-product impact
Agent AI extends beyond traditional chat forms; its capabilities can be integrated into other product lines to automate and optimize their workflows. For instance, in hotel bookings, Agent AI can filter and recommend hotel options based on user preferences and the context of their upcoming trip.”
A shift in user perception towards AI
We’ve observed an increasing number of users who interact with Agent AI as if they were speaking with a human, often without realizing or caring whether it’s a bot.
Leave a Reply
You must be logged in to post a comment.