The Product Designer’s Guide to Conversational Commerce
Trevor McNaughton
2597

The difficult part for me with chatbots is that while there are services, like init.ai which say that they understand natural language, they still will depend on predetermined responses, which don’t feel like they were written by anything but a bot.

I’ve seen and interacted with some great conversational interfaces, but at no point did they suspend belief that it was anything but a staged interaction. People are pretty good at smelling out the canned response, and will respond to the chatbot accordingly. I think your conversations in the article are cool and all, and I’m sure that your platform will accommodate them, but are you really seeing people respond with the tone that you showed in your examples?

For example, in a few of your messages, the user chooses an emoji along with text… to be blunt, who would do that? I’m not saying who would use emojis in a conversation, but who would use one when talking to a machine? How would you give your users the level of trust with your platform that they could use language like that and not confuse the bot?

All it would take was one “I’m sorry but I didn’t understand that” or something similar to completely undermine the whole conversation and the user’s perception of the recipient’s ability to understand that they are saying. Also, phrases like “you can ask me” will put you in a similar pickle. After reading that, your users will know without a shadow of a doubt that they are not talking to a person. It will shape the tone, length, and phrasing of all of their responses afterward.

I’m a big believer that the conversational interface is on the horizon, but right now there’s still an understanding on syntax that is necessary, and I can see that making each experience less than the real conversations that you describe. I know that if I’m talking to a real person about getting tacos and then suddenly change context, the person will be able to change track with me. I find myself having conversations regularly in which I am simultaneously discussing where to go to lunch, what the status of my work is, and what the weather is like with no effort. How could I have such a conversation with a machine without getting countless error messages?

Not saying this is a discredit to your service, and I realize that this is a bit of a rant, but it’s the big barrier in my mind when looking at this type of interface. How can anyone possibly train their chatbot to be able to consume as much random bits of data as we humans produce in our daily conversations? How can we design an experience with a chat interface that enables users to talk the way they would to a person? How to we teach them that they can say anything and the bot will understand it? In other words, how do we make it an actual conversation?

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.