Chatbots are harder than you think (video)


After watching the video above, you’re probably thinking: Sounds easy. Too easy . . . and you’d be right! Chatbots are the shallow end of the AI pool but never underestimate how quickly a simple conversation can go off the rails. This tool could be a great start but as you suspect there are several things to consider.

  1. Start with a clear use case – Use cases are essentially stories that you write before you begin coding to help spell out the majority of interactions the system will have: step-by-step. (User asks a question. System responds.) These are then prioritized and developers work on the most essential ones.
  2. Communication (not just information) is the goal – If you were asking someone for directions to their current location, they would send a map with their location highlighted. If someone were asking you what a piece of art looked like, they would send you a picture. Your bot will need to optimize the media you share to aid in communication.
  3. Cadence counts – If you watch the Taco Bell Tacobot interaction below, you’ll quickly notice the inhuman speed with which the bot replies. It should! It’s not human. But you may also notice that it makes you feel uneasy. Personally, it makes me nervous. Many times when you get a metric ton of content back at you, it’s because you’re in the midst of an argument/serious conversation. You’ll need to balance out the need to get your customer(s) the right information in a concise manner in a way that feels quick but not overwhelming. (We’re not trying to trick customers into thinking the bot is a human – like the sounds of keyboard strokes when dealing with an automated call center system. We’re simply communicating with the right cadence.  
  4. Context counts – “Is there a branch near me?” It’s easy to imagine your bot being asked that question. Also, imagine that question being asked when your branch is closed. That should definitely change the answer. Context is essential.
  5. Plan for sustained “flexible” conversations – Imagine you’re bot is mid-conversation with a user transferring funds from one account to another and then the user asks how much money they have in their account. This is a completely natural inquiry but if you don’t plan for it with your ‘transfer funds bot’, you’re toast.
  6. Emotional state of the user – If a user’s flight has just been cancelled, it’s safe to assume, they’re not happy. It’s probably wise to send them to a live customer service agent instead of a bot. That said, there is a company called SenSay that has technology that claims to be able to tell the emotional state of a user.
  7. The experience is the brand – There is no need to mention your brand in this interaction. They know who they’re talking to and you should spend more time making sure the experience delivers on the brand promise.

The foundation of this list came from @setlinger, whom you should follow, and I’ve added/changed where I saw fit.

By Michael Myers