Overview
ChatGTP has put a whole knew spin on the popularity of AI, but people still don’t trust robots. We like the connivence, but we don’t trust them thinking they’re going to take our jobs. That’s pretty clear in that only 1 percent of users are willing to open a launcher for a chat window and even if they do that only 30 percent of the 1 percent continue to a conversation with a bot, most skipping over to requesting a human; how many times have you shouted “AGENT!” at the phone? This project was an effort to shift the mindset of our team. My goal here is to identify the value of taking a step back and with theory apply user data to shape business requirements, instead of business requirements shaping user experience.
Opportunity
An introductory epic, technical scope is limited for any contextual APIs. Additionally, the foundation of what we provide to our customers can only take them so far. The written voice of the virtual assistant to communicate tone and purpose is out of our hands. We can only provide the platform and support education.
Business Goals
Improve industry standard conversion rates of opening a chat launcher from 1% to 3%.
Design Goals
Identify user gaps in trust and discoverability to provide valuable support from a digital assistant.
Research
I started in sci-fi. Humans have been dreaming about robots since the Czech playwright, Karel Čapek, invented the word Robot in his play Rossum's Universal Robots, in 1920. But why should creators of AI today care about the fantasies of Sci-Fi and fantasy? How our artistic partners in story telling have envisioned AI has captured the hearts and minds of our users. When they watch HER, they begin to believe we’re on the brink of developing relationships, as did an individual in Japan who has symbolically married cyber celebrity Hatsune Miku. By looking at movies I put together a list of good vs evil characteristics. When we start to overlay the take aways from movies with system-human relationship models we see symmetry.
Bad robot
good Robot
Good Characteristics
Collaborative team member
Purpose provided in introductions
Permission required
Providing awareness (ie. loading information)
Multiple format response (audio, visual, text)
Awareness of PII
Data supporting recommendations
Conversational tone
Evil Characteristics
Autonomous worker
Unsolicited advice
Action without permission
Bogarting human choice
Driving agenda without human overrides
Source logic undefined
Data without recommendation
Relational development model
Mark Knapp is a teaching professor at the University of Texas and is known for his works in nonverbal communication research. Knapp's relational development model unpacks the build of trust over time, from the initial introduction to a fully bonded emotional connection. Designs that carry the user through the story arch of relational building with the AI that supports their work will sustain greater adoption. Often support tools are overlooked because they jump from interaction to an integrated approach before ever being introduced, or they exhibit the evil characteristics reviewed earlier.
Purpose
Reason to engage with the system in the first place.
This can grow and change over time.
Value
How the system tangibly improves the users’ life.
Trust
Willingness to engage in a relationship and disclose personal feelings and/or information. Starts small at the beginning and grows over time.
Healthy relationship
Confident user resulting from the combination of purpose, value and trust.
Knapp Relational Model
Building blocks of healthy relationship
Solution
So here’s the big question: How do we improve the 1% conversion rate of users clicking on a launcher and what we know about human’s biases towards evil robots? I started by overlapping a high level journey map of the end user, Cade, and Knapp’s steps in building a relationship.
Journey map above, Knapp below. Potential touch points to elicit steps in the center.
Then I used a metaphor - walking down street filled with shops. There are lots of things to catch one’s eye on the street. People shouting at you to come in like sign throwers or fundraising clip-board jockeys. They’re pretty in your face and people would rather avoid than interact. Then there’s the nice widow display with a table outside so you can sample before committing. Sandwiched in between are the blank storefronts that are uninspiring or the thoughtful bench but again there’s no company interaction.
Now when we look at the browser window we have different elements on the page, how does the launcher find any attention, and where would it’s presence provide purpose and value to Cade? Using Gutenburg’s F and Z reading patterns the top right corner of the screen has the most viewing traffic. Groups like Nielsen Norman Group have identified that users expect launchers in the bottom right corner instead. However, my UX researcher and I ask: if that’s the case and that’s the correct place to put it, why doesn’t anyone click on them? We plan to find out more.
Street filled with shops
Browser window
Impact of mindset pivot
By going through this process with the team I was able to pull the group back and look at the big picture to identify the user ‘whys’ we were solving for. Product management and development were already focused on components which were pushing us in an industry standard approach. However, if the industry standard is 1% conversion rate and we are trying to change that, we had to look at launchers from a new angle. This foundational mindset shift I placed in the team lead to changes to our acceptance criteria, release priorities and resulted in a 200% increase in conversion.