Taking on the role of lead copywriter and content strategist on the project, we had an early opportunity to create a chatbot. Successful bots often focus on a core feature, but our task was to create a proof-of-concept bot on behalf of a provincial lottery corporation that enabled users to buy tickets, check estimated jackpots, learn about each lottery game offered, and — of course — let you know if you won.
To start, I mapped out a decision tree that supported the conversational logic for the bot. Relying on AI to process natural language is, to put it lightly, an inconsistent process. Ensuring this flow remained as flat as possible was essential, as every branch adds another opportunity for the bot to fail. I was ready for that though, designing contextually relevant error messages into every branch of the tree. These error messages would guide you back into the flow at relevant positions. After all, why do we fall sir? So that we can learn to pick ourselves up.
With a decision tree laid out, I was able to tackle actually writing variants for each message the bot sends. Adapting to the context of these new platforms proved to be essential. When texting with a bot on Facebook Messenger, you can easily scan a message for important details. This gives you the freedom to add a bit of personality to the bot. However, when we took the bot to a voice-powered Google Home, an extra few leading words to create suspense just feels like your time is being wasted by a robot.
The most exciting part of the project was putting the prototype in front of real users. From users treating the chatbot like a real human (asking “How are you today?”) to treating it like a command line interface (typing “/help” when they were stuck,) user feedback revealed how people would react to an entirely new way of communicating.