Ferris was a text messaging service to help people break from routine and discover new experiences in their city. I launched a prototype in a weekend and grew the service to 1,000 monthly active users.
It is the experiences we live, not the items we possess, that make us happy.
There are plenty of websites and apps out there that help us stay connected to the events happening in our city, but all take the same form: a wall of options with categories, filters, and reviews to help us refine our search. In reality, when presented with so many choices, the paradox of choice sets in and it becomes more difficult to choose.
The one existing solution that works well is the adventurous friend that some of us are fortunate to have. The adventurous friend is comfortable with the uncomfortable; they break out into the world and inspire us to go with them.
Ferris aimed to be that adventurous friend, as a service.
Ferris was started during the early wave of the bot-craze. The computing power of our phones can be moved further to the background, and a simple command (such as a text message) could initiate a service.
Text messaging felt like a natural medium for the adventurous friend as a service (AFAAS...?).
I setup a Google voice phone number and posted about the project in reddit.com/r/Boston. Thirty people texted Ferris to find events in Boston that day. I manually fielded every conversation to understand how to craft a conversation flow that might be programmatically repeated.
Over the following few months, I went through a rapid process of tweaking and adjusting the service as it organically grew to 1,000 monthly active users.
I built out a Rails app to host a database of event cards, which organized basic information about an upcoming event. When Ferris (or the human behind the curtain) was ready to make a recommendation, a link to one of these cards was texted to the user.
The goal was to finally phase out the manual process of fielding requests, and to get Ferris fully operating as a bot.
A friend of mine started helping with the project. He built out a tool to scrape events from Facebook to populate our database, and built in some simple natural language processing to pick out keywords in text messages and reply with relevant events.
Unfortunately, we couldn't get the bot to a point where it was intelligent enough to match the degree of personalization offered by a human operator.
After a few months of manually fielding messages and prototyping dozens of alternative services, I shut the service down.