Overview
ASAPP offers GenAI (generative AI) services to large enterprises (including major US airlines) to use within their current CX stack. Over the course of 2022, I worked on our Virtual Agent Tooling Platform(VA) to empower contact center managers to create timely and personalized experiences for their customers.
For a detailed presentation of this project please check out the Figma link below ⬇️
What does a virtual agent do?
Simply put our virtual agent would turn regular conversation into machine readable data. After which this data is categorized and mapped to certain actions that will be communicated to customers interacting with VA’s bot.
The challenge
All existing customer messaging campaigns and flows in our Virtual Agent platform were coded implementations. These flows were essentially just JSON files. Every time a customer wanted to publish a new flow or modify an existing one, an Integration Management team member had to build, test and deploy it. This process on average took 12-15 working days.
Goal
We wanted to offer our customers a visual flow builder that would allow them to quickly test and deploy their desired flows. The end goal being that they will deploy many more flows addressing various customer needs which would result in a decrease in agent escalation.
My Role
I led end-to-end design efforts including scope definition for MVP. Due to strict timelines, I conducted a lot of guerrilla testing sessions (🐇💨) to iterate and validate designs rapidly.
I was also responsible for our training documentation and on-site training with 2 of our clients. (American Airlines & JetBlue).
Initial research
Being completely new to the gen AI space I set out 3 primary research goals for myself to better familiarize myself with your offerings.
Reviewed every flow for each of our main customer accounts (~250-700)
Interviews with our Integration Managers (3 👩💻)
Working session with Eng. to understand each component (node) within a flow
Creating a visual flow builder
I primarily relied on early usability testing and feedback sessions internally/externally to validate the overall design direction quickly.
I used Figjam (shared with our customers as well) to quickly come to consensus on the overall anatomy and structure of how nodes should be displayed and connected to each other.
Quick & dirty usability testing
All concepts were tested as low-fidelity prototypes with our main customers for feedback and direction.
Polish & finalized mocks
In order to move rapidly the finalized designs leveraged our existing design system used in other platforms. This reduced the overhead both on design and also the implementation
Training material
Alongside my product manager, I contributed to creating training documentation and conducting on-site training for our customers.
Impact
AA more than doubled the number of flows they had since launch. (~ 1400 flows)
JetBlue saw a ~5% drop in number of agent escalations.