Meet Scarlett

The goal of this project was to create an interactive, text input digital assistant for Safelite’s Help Center by using StoryFile’s Conversa software. The idea was to have customers be able to ask their questions to a digital human, and have a better, more positive experience with the Safelite Help Center. 

As the digital assistant (Scarlett) was to be on the landing page, we had to consider: Scarlett would catch a user’s attention immediately, as she was placed at a focal point. How would that impact the user's experience?

Safelite intended to have Scarlett speak all the information that was available. How would that information be presented through a digital assistant, versus a regular text box? What would sound natural or what would sound more stilted? Additionally, working more as a SME on this project, how do we give a crash course on conversational AI so a client can eventually manage their own product? 

For this particular project, I taught Safelite how to implement a proprietary platform, along with guiding them on how to restructure their FAQ content for conversational AI. As Conversa from StoryFile enables individuals and enterprises to create and manage their own multi-modal experiences, it was important to me to also cover some guiding principles for conversation design and the end-user experience to keep in mind while building their product. This was accomplished through a series of training sessions that I led.

In these sessions, we discussed how much of the content from the FAQ was to be included. Based on those conversations, Safelite rolled out content in phases. I also discussed with them the importance of pacing information to users: if users are overwhelmed by the amount of information they’re receiving, or not getting to their goal fast enough, they risked losing out users on interactions.

During some initial interactions we also realized that while our intention was to have Scarlett as a digital assistant that gave information buried in the Help Center, we quickly realized that many users treated Scarlett as a help desk (something I had also noticed in other client’s projects). So sometimes, users would type in something like, “My windshield is broken and my case number is [123456] and I live in [Somewhere State] what updates are there.” To field these queries, I proposed two solutions: a) implement some classifiers to easily sort through information and b) redirect users to a different part of the website, or support, etc.

Additionally, there was little to no guidance for the user to know what exactly they were looking at. While this leaves a clean-looking UI, it also risks establishing a baseline for the user experience. Without scaffolding, users could begin interacting with preconceived notions and if those notions weren’t met, left with an unsatisfactory experience. I suggested adding a quick intro video that establishes the role of this assistant, and with an open-ended prompt to encourage someone to type in the response. While this can be risky to do, as this can leave things too open-ended for users, Scarlett was slowly being updated with all the content available in the Help Center, and the client was comfortable with continuing uploading content to eventually capture all of the user’s queries, along with the appropriate error-handling and redirecting.