Before you build your bot: what it takes to make a successful chatbot

In an ever-expanding sea of chatbots, how do you stand out from the crowd?

chatbots chatbot bot
Thinkstock

With hundreds of new bots popping up daily, it seems everyone is exploring chatbot development. In fact, the chatbot market size is estimated to grow from $703.3 million in 2016 to $3.2 billion by 2021. In an ever-expanding sea of chatbots, how do you stand out from the crowd? With 49.4 percent of people saying they would rather contact a business through messaging over phone, building a chatbot should be on your to-do list. Here are some tips and tricks to make your chatbot successful.

1. Chatty chatbots are best

The best chatbots sound and read human. People make the occasional typo or spelling mistake—and so should your chatbot. If every response is typed flawlessly, it could come across as robotic and too structured. Give your bot the freedom to be a little casual, code in the occasional typo—and always remember to keep it conversational.  The best bots learn from the conversation and weave those learnings into the bot/human exchange.

For example, Georgia Tech’s teaching assistant chatbot, Jill Watson, built on IBM’s Watson Platform, learned from dozens of conversations with graduate students. With four semesters worth of data and 40,000 questions and answers as its backbone, the bot read forums and studied how the students used inside jokes and jargon in conversation. Eventually, the bot filters these jokes into the conversation. By the end of the semester, students thought Jill Watson was an actual person, not a chatbot. The idea isn’t to deceive your users, but instead to ensure your bot comes across as conversational and easy to interact with.  

2. Memories drive conversations

The best bots personalize the conversation and help people complete a task, thanks to artificial intelligence and machine learning like IBM Watson and other tools. Context storage is key. Your bot should remember who the user is and key facts from previous user engagements. Context storage is easy to set up using a database connected to a back-end server. For example, if you ask a bot to give you restaurant recommendations, the bot could suggest a particular place based on your past preferences. When you ask the bot, “Why did you recommend this restaurant?” the bot should be able to reference your previous restaurant visits and preferences. The human memory capacity is 2.5 petabytes (2.5 million gigabytes), and soon enough bots will catch up.

3. Build a fallout mechanism

In the current chatbot environment, around 70 percent of all human responses go misunderstood. For example, if a user were to ask your bot a question outside the bot’s subject matter expertise, instead of responding with “I don’t know” or “I don’t understand,” the bot should have a response similar to “I don’t have that information, click here to speak to a person.” The chatbot needs to be able to respond; the user isn’t going to always spell correctly, or use perfect grammar. Many bots don’t account for this and lose the user purely because the wrong type of “there” is used.  It’s important for you to not only have this fallout mechanism in place, but to also be able to learn from it. The bot can return this data, and allow the programmer to correct it for the next interaction. Contentedly, a large percentage of customer service is considered to be 80 to 95 percent precise after only 300 human annotations.  

4. Consider the user interface before you build your bot

While you might know what you want your bot to do, before you even begin programming, think about how you plan to deploy the bot. First, you must make the distinction whether it is going to be a chatbot or a voicebot. If it is going to be a chatbot, consider whether that’s Facebook Messenger, Slack, or through the web; each platform has unique characteristics and generally serves different audiences. The medium influences how it’s used, so it’s important to understand your users and where they are most likely to use your bot. You’re going to get the most engagement with your chatbot if it is through a channel your demographic is already using.

5. Chatbots are more than just customer service engines

While public perception is that the majority of chatbot development has been in the area of customer service, the potential for new applications is endless. Chatbots are currently used for habit formation, navigating the US immigration system, and even in fundraising efforts. There is a bot named Yeshi meant to be a young girl living in Ethiopia that you can interact with through Messenger to learn more about the water crisis. The bot simulates what it is like to have to walk for over two hours just to find clean water and responds with images, GIFs, and videos detailing the daily experience.

With thousands of bots deploying each day, chatbots will continue to expand and find their way into all industries with new use cases constantly being discovered. Bots are ubiquitous, and with little barrier to entry, you could be building a bot that completely changes the way people do things, in no time.

This article is published as part of the IDG Contributor Network. Want to Join?