About the project

As we all know, Human-Centered Design (HCD) is a problem-solving framework that is grounded in empathy and understanding. I am a big fan of it because I believe it could help make systems and products more responsive to the people, or the customers, who use them. 

During my design period, I tried to apply HCD methodology as much as I can, even with a tight deadline, by conducting rigorous qualitative research to deeply understand the needs, insights, and emotions of my audience. 

The Design Process

The overall design process I used could be summarized as the Double Diamond above. The process is cyclical. People and their needs are dynamic and changing and so our solutions are dynamic and changing. During the design process, once a design solution is launched, I measure its effectiveness against initial and intended aims, and then I continually tweak it, thus improving the solution over time. 

Understanding agent types, behaviors, and working process

Who am I designing for and what are their behaviors are the main questions I ask before kicking off any of the designs. In this case, my audience for design is the Expedia global contact center agents. Based on the discovery I found that there were a lot of complexities that needed to be considered when designing for our agents such as agent tier, experience, and location.

At Expedia, we have

and they

User Research

Understand agents’ needs and pain points 

I conducted multiple workshops with flight experts and stakeholders to understand the agents’ needs and pain points for the current flight change process. This helped me understand more about the complex flight scenarios, and was also very helpful with identifying the use cases. 

After gathering mostly qualitative data, I put together a journey map to illustrate the actions, touchpoints, required tools, feelings for agents doing a flight exchange. This journey map was share-out with a wider team including product managers and content strategists to gain a better understanding and alignment of how the ideal future chat-based flight exchange process could be, and where we could build the eligibility check and logic points.

Image: agent journey map for flight change

Insights

I was not too surprised, since the current CRM applications were developed in approximately 2000, and I know agents eager to have a brand new streamlined experience to make their life easier. According to the data from the data analysis team, Expedia needs this to be designed to increase agent engagement resulting in reduced operational costs, and it also meets our organization’s business goal to retire the old tool Voyager Flight. And that’s my mission. 

Partially automated flight change:

Imagine the possibilities:

Then based on the insights from user research, I started to explore the ideal-to-have features we could include in the virtual agent experience for the agent:

Here to show 3 scenarios:

Scenario 1: Flight Change with a less than all travelers and automated waiver

Scenario 2: Flight Change with unparsed fare rules

Scenario 3: Flight Change – Virtual Agent needs waiver info

The Robust Card Design

– Working with a content strategist & engineers along the way

After ensuring the experience makes sense and defining where to use the robust cards, I then started the robust card design. The goal is to use these inline cards to allow the agent to change the flights. Due to the NLU limitation, we have to use robust cards for:

• Booking selection
• Time/ Date/ Location selection
• Flight search
• Search results
• Review new flight
• Payment
• Updated new flight card
• Fare rules
• Waiver card
• PNR

The stories behind the cards

At the same time, me and my content strategist worked together to apply the best practice for both design and content, meanwhile delivering a detailed UX conversation workflow to identify the ideal user experience. Since this is the first agent project, the flow has gone to rounds of design reviews with senior leadership, and I also got the chance to work with content leadership to define the guidance, and voice & tone too. 

The conversation workflow

The flight change for agents follows this very simple and basic pattern:

• Select the flights and travelers you want to change
• Input the new criteria (dates, times, origins, destinations)
• Select the flight that best suits the customer’ needs
• Pay for the flight in the customer’s preferred form of payment, if payment is necessary
• Receive a confirmation that the flight has successfully been changed and the new itinerary

North Star – Designing with NLU first approach

Always put aside the limitations and imagine the best North Star experience. The North Star design idea is to capture and parse the customer’s needs, intents, insights, flight info, airline change policies, and fare rules intelligently with the help of artificial intelligence, and natural language understanding.

Let’s first take a look at what the original workflow looks like.

Define the Phase 1 workflow – what we could deliver based on the tech limitations

After showing the North Star workflow design with engineers, we learned the NLU technology is not strong enough to allow the virtual agent to watch and learn the customer inputs, but I believe eventually we will get there! The next step is to design a workflow that we could launch for phase 1. The Phase 1 workflow has also gone through rounds of design reviews with senior leadership, stakeholders, and engineers.

Let’s first take a look at what the Phase 1 workflow looks like.

Phase 1 partially automated workflow

Before starting the partially automated workflow design, I have so many questions that needed to be answered.

Scenario 1: unparsed fare rules

When automation is not available (system errors, fare rules cannot be parsed, etc), what will be the agent experience?

Based on the exploration, the ideal experience is to display the fare rules for agents and have the agent put in the change fee amounts. In this way, the agent won’t need to either checking generic fare rules in the knowledge base, calling airline companies, or leaving and redoing the task in GDS. It could help the agent to save time and simplify the process.

Scenario 2: Ticket stats is expired

When the system is not working, or the ticket stats is expired, the system should lead the agent to GDS. This is a must-have graceful failure case, and this is what all the global agents need to do at the present stage.

However, what could we do to enhance the GDS hands-off experience? 

Two things I pushed here are:

• Checking with the flight experts how does the agent determine which GDS to use under what circumstances? If we could define the scenarios, we could then show the agent which GDS to use in the graceful failure message.
• Can we directly hand off the case of the expired ticket to the T3 agents?

Reasons based on the user research findings:

From the early user research, I learned there is not only one GDS out there. Global distribution system (GDS) is a computerized reservation system for reserving airline seats, hotel rooms, rental cars, and other travel-related items. Amadeus, Sabre and Travelport, Apollo and Galileo are all global distribution systems. In the phase we refined the scopes, my pm and Service Delivery team defined we only considering Sabre, Amadeus, and NDC inside the scope for Phase 1 delivery.

Another finding from the user research, when I conducted the interview with agents, one of the agents mentioned that only T3 agents could read the GDS command line since T1 and T2 agents have different agent training which doesn’t include how to use GDS.

Scenario 3: Waivers

When natural disasters occur, each airline will come up with related policies, which we called Flex Policy. When the ticket is eligible for waivers, and the Flex Policy applies, in the ideal process, the virtual agent could proactively know there is a waiver and could intelligently show the date restrictions before capturing any further change intent. However, when the system is not working, a fallback option could be displaying the designs for the agent to apply a waiver code. Then agents won’t need to go to the knowledge base or airline website to find the waiver code.

For this complex project, we have to work on many use cases, think about the messages based on the airline policies, the system fall-back, graceful failures, agents’ behaviors, and legal comments. Due to the NDA, I cannot show too many details here. But, have a quick glance at the final messages and feature designs.

The conversation design showcase

Thoughts to share

Why focus on being systematic and platform-minded when doing the conversation design? 

• We’re creating cohesive design solutions that aid users in comprehension, trust, and adoption as customers return, again and again.
• Unique “snowflake” solutions are harder to maintain internally. Uniformity is more scalable.
• Systems thinking allows us to work smarter, so we don’t have to duplicate work and ask the same questions others already have. 
• Valuable for the ‘Localization’.

The Findings: High-level takeaways

• The overall process was perceived as easy and friendly
• All participants were able to complete the process easily with minimum guidance or confusion
• Most of the participants were able to interact with the UI patterns without confusion
• PNR Details are not easy to understand, would be great to only show that for T2 and T3 agent. (T1 agents do not necessary read the PNR details and they were not got trained for GDS
• Agents loved the new design could stick to some features in Voyager Classic & Voyager Flights. (Note taken session was missing from Voyager Classic, and agents were eager for having this feature.)
• Agents wanted ‘city’ instead of airport code
• All the chat participants (13/13) loved the North Star idea of sharing content (policies, flight details, confirmation messages, etc.) and felt comfortable about the level of help from VA

The launch 

Fully launched

The product launched at the end of 2019, and after that, we then did a fast-follow based on the user testing findings in Jan 2020. (eg. adding airport codes to the carousel, etc. )

Fast-follow based on the user testing feedbacks

Creating training materials for agents

After the product launched, I later partnered with an instructional designer to deliver training materials for call center agent training. Here are some of them, click to play the videos if you find them interesting!

There is no perfect design, only the better design

Be open to the failure and feedback, and think of how to make it better.

Cherish all the friendships and relationships

As designers, we have the mission to make this world a better place. Instead of work, one thing that we can do is to build relationships because as time goes by, everyone will realize that work is all about relationships.