The Burner team came to the Voxable consultancy because they wanted to create a chatbot to connect Burner’s disposable smartphone numbers to helpful services. After we discussed the team’s ideas and user needs, we uncovered the online dating use case.
We needed to explore which dating use case made the most sense for Burner’s users, the platform, as well as the conversational AI capabilities available. Finally, we needed to deliver an engaging chatbot that brought Burner attention and engagement from their existing users and also attracted new users.
With the help of the Burner team, and my research into Burner app store user reviews, we identified an interesting user segment: people leveraging Burner’s disposable phone numbers as identity protection while dating through apps and websites.
I was responsible for the design strategy and execution of the Ghostbot branding, in-app onboarding screens, and conversation design. I worked with the software engineer to define and implement the app onboarding experience, chatbot experience, and the natural language understanding (NLU) model that powered the chatbot. I also collaborated with a writer/information architect in the creation of chatbot training data.
While we explored one of the most popular use cases for the Burner app online dating, we discovered the user segment skewed toward marginalized populations who are more at risk of harassment and violence while dating. For this use case, people primarily leverage Burner as a privacy layer between themselves and their dates. My research discovered that women experience a disproportionate amount of online abuse and aggressive texting in short-lived relationships. I collected and categorized the abusive conversations women documented and posted to sites like Tinder Nightmares and Bye Felipe.
What we found is that the abusive conversations seemed to happen no matter how women respond to dates. We reached out to women who were active in the online dating scene and heard similar reports of men suddenly becoming abusive when the relationship dissolved or some other perceived slight occurred.
With the Burner team, we landed on the idea to create a chatbot that could ‘handle’ the abusive conversations. Users could switch on Ghostbot when their dates became abusive. The chatbot would be trained to respond to the harassment, and the user wouldn’t have to deal with the emotional toll of having to engage in the abuse. The bot could even automatically block users who reach a certain threshold of abusive language.
With women users as the primary focus, I developed the branding and messaging for Ghostbot. I created the Ghostbot logo and branding used in the application and worked with a writer to shape the bot messaging within the chatbot.
I designed the natural language understanding (NLU) to respond to various incoming text messages and built the conversational intelligence to respond to texts, even the most aggressive ones. I created training data and sample dialogues to train the NLU model built in Dialogflow.
I designed the app onboarding experience for activating Ghostbot on the Burner platform, which included creating a user flow and UI screens.
I designed the mobile-optimized landing page as well as the in-app chatbot onboarding screens. I created the design specs and assets and worked with the developer to ensure it worked as expected. We leveraged Usertesting.com to test and improve the onboarding experience. I managed the testing set-up and delivered results in the form of highlight videos and design recommendations to iterate on the solution.
Ghostbot had an incredibly successful launch that was picked up by dozens of news outlets. I spoke to reporters about the research we found, our design approach, and how we built the AI backend. The chatbot was subsequently enabled by thousands of Burner users with a significant proportion being new users. The Burner team was happy enough to work with us again on the Hostbot project.
As one of the first large conversational AI projects we took on at Voxable, this one helped me learn a lot about the process of building applications that leverage conversational AI. In particular, which user interface elements are important in a chat interaction, the amount of training data required to create a solid NLU model, and the unpredictable nature of conversational interaction.
