Banking Virtual Assistant that learns
CLIENT
Kasisto, Inc.
What is KAI?
TEAM
VP of Product: George Trad
Chief Marketing Officer: Stephen M Epstein
Director of UX: Francesco Nasso
UX Research: Mar Alexanian
ROLE
UI/UX Designer, with focus on Visual Design and Animation
PLATFORMS
Web, Mobile
Overview
Kasisto needed a vision. One on a low budget and limited timeline. I was contracted in this role and given no more than 20 hours per week to put forward a concept for VA Avatar states, animate them, deliver wireframes and visual designs of both low and high fidelity.
The product’s engagement metrics had room for improvement. So, how the Virtual Assistant looks, feels, behaves was to evolve in order to demonstrate immediate value was crucial.
Generally the virtual assistants of the day and the ones I personally looked at were: Alexa (Amazon), Siri (Apple), Cortana, Bixby, Lyra, Viv, Ozlo. And specifically for banking: Clari (TD Bank) and Stacy (Standard Chartered).
VAs are meant to replace human conversation via auditory and textual means. While chatbots are server or company oriented, a VA is user/customer oriented. It is an application program that understands natural language voice commands and completes tasks for the customer. VAs save money in sales, since they are an addition to the website. They help customers troubleshoot, work with discount codes, loyalty programs and solve almost 70% of other customer issues without needing human intervention.
Create the visual design and set the stage for the user experience of an AI Virtual Assistant, instead of just another banking chatbot. Design a data-driven Avatar, that is the cause of engagement for users to take action within the system.
Challenge
When I started thinking about what this “character” would be at its core, I immediately thought about mirroring a personality. My thoughts were in the direction of a hairstyle, clothes, and accessories, all customizable. But was it a man or a woman, are they elderly and wise, are they standing or sitting down? In terms of emotions, perhaps I could design a set: one friendly, one serious/neutral? I thought of forming the character after people like Warren Buffet or Christine Lagarde, someone who would immediately be associated with finance. And secondly, since the name of the company, Kasisto, translates to Cashier or Treasurer from Esperanto, I landed on an idea about a coin purse named Pursey.
Thought Process
Kasisto’s logo broken apart, to make a virtual assistant avatar.
Some very rough iterations of Pursey.
A week into the project Kasisto’s banking virtual assistant remained illusive. I even experimented with breaking the company’s logo apart and having it animate, which in retrospect, was the right direction for what was to come. After my first presentation I received a most valuable feedback from Kasisto's Director of Marketing and some constraints that truly helped further my work.
The VA was to be named KAI, after Kasisto Artificial Intelligence. It was to be a genderless objectless entity that represents a higher promise beyond just a persona. It was to be an experience. In other words, what Kasisto brings to the table is a model driven by data, instead of a conversational persona. It could abstract and therefore lend itself well to various branding. It has movement to it, as well as states. As in: state of Alert, state of Thinking, state of Calculating, Idleness, even Regret or Satisfaction (Happiness).
Initial ideas for different Avatar states. These are starting to get more refined as I work on them, showcase and gather feedback with every iteration.
An early idea about the User Inactivity state: With time, KAI turns itself into a sunrise - sunset scene and night sky with the moon appearing. Because KAI knows the birthday of the particular user, the star constellation in the night sky turns into their star sign. The storyboard scene can be used for other special occasions as well. The animation is smooth and the music is relaxing.
How we tested?
Test users were shown static images of KAI avatar states and asked to match the image to a description.
For any avatar they couldn’t match to a description, test users were to select: “I’m not sure” and provide feedback in free form.
Total number of participants: 27
Process
KAI in action. A first iteration wireframe showcasing Quick Access mode, for mobile.
KAI Alert notification
KAI Alert notification, in stressed state.
KAI - in locked context conversation
Outcome
The larger majority of users were confused by the avatar states.
They expressed the states were too ambiguous and unclear as to what they were supposed to be. Overall, only a few avatar states seemed to match the intended behavior.
The majority of testers enjoyed the process of trying to match the avatar states with the appropriate image and expressed positivity about having a multitude of designs to choose from was an engaging activity.
We knew we had to add motion to the design of KAI so that it would simply become more accessible and a lot more useful.
KAI states animated. They were turned into JSON files and handed off for further development work.
KAI’s greeting and introduction state.
Another iteration of Greeting and Introduction state.
Let’s chat.
Notification.
Happy.
Thank You.
Idle KAI.
Survey.
Alert.
I’m sorry.
Calculating.
User inactivity.