AI IRL Podcast Episode 1: At the Intersection of UX and AI
Ultimately a product is only as good as its usability. The UX/UI of artificial intelligence is vitally important. With so many directions to explore, what are the areas that actually lead to real end-user improvement?
Kat’s a Senior UX Designer with over a decade UX experience, contracting for Levi’s, Charles Schwab, Blue Bottle, EnChroma, and several others.
On this episode, she gave us an inside peek into the world of UX design and introduced us to AVA: Asurion’s Intelligent Insurance Chatbot
Asurion is an insurance provider, primarily for mobile devices. Kat works with the development team on the UX side to create an end-to-end claims process built on AI. That’s AVA.
“If our customer had any option in the world, what would they to choose?” — Kat Dykes
AVA doesn’t just file claims – the bot can handle a range of complicated issues, including phone repairs. If you are insured by Asurion and your phone screen cracks, for example, AVA can help you schedule a technician can come to your home or office to repair your device.
Designing a UX for NLP Is A Whole Different Dance
AVA is built on a complex natural language processing (NLP). The experience Kat brought with her to Asurion was undoubtedly helpful, but she quickly realized AVA presented a refreshing challenge.
Her last project was at Levi’s, where she was working with responsive websites. Mostly, consumer interactions from click responses, going from one page to another, and piecemeal components. But the idea behind AVA (and many customer support bots) centers around the creation of a fluid conversation between interactions.
“For the past year, we’ve been working with and furthering our understanding of conversational design.” She said it’s challenging, refreshing, and complex, all at the same time.
Building on NLP over responsive websites, is an entirely new type of dance.
In NLP, You Must Resist the Button-Urge
Coming from a responsive background and a designer’s heart, Kat and other designers have to resist the urge to constantly create traditional carousels or buttons.
“All of us constantly want to insert buttons and all sorts of easy-to-interact-with options, but we really need to step back and understand if that’s actually the best experience for the end-user.”
“As designers, we constantly want to insert buttons, but is a quick button really serving the user as best we can?” — Kat Dykes
Often, AVA uncovers issues and understands the customer by not giving the carousel options upfront. If you immediately fall to a list of options, you risk influencing the customer’s thinking before you’ve understood their problem. But by initiating and allowing a conversation, sophisticated AI can determine, with a high level of accuracy, what underlying problem the customer is trying to solve.
And it’s working phenomenally well. Currently, Asurian is seeing success percentages above 95%.
Marry Conversational Flow With Traditional Options
Kat’s team has determined a crucial key in understanding how to maximize AI with the best UX: a marriage between conversational and traditional AI methods.
For example, when a customer is trying to schedule a repairman, the AI will typically start with a conversation — asking the customer what they need and what they are trying to accomplish.
Once AVA determines what the customer’s trying to accomplish, they will continue forward, even if there is one or two pieces of information missing. By using machine learning, AVA can often determine a “best guess” on what that layer of missing data is. If it’s unsure, or needs clarification, it’s at that point, AVA may present a carousel.
Finding out whether conversational AI or a more traditional approach is best can be a challenge. But it’s necessary when the ultimate goal is usability and ease for the customer.
After many A/B tests to determine which option is better, they settled on a marriage of the two.
Last Tip: Our Parsers Are More Powerful Than We Think
“Our parsers are far more powerful than we think.” As Kat continues to push the boundaries of development with their chatbots, she learns that their parsers are well-designed, and increasingly able to handle complex NLP.
What does the future hold?