Artificial Intelligence

How to Build a Better Chatbot

Six best practices for building better bots

By Suhas Uliyar, Vice President of Bots, AI, and Mobile Product Management, Oracle

Spring 2018

Not all chatbots are created equal. Here are some best practices for creating chatbots that drive greater customer engagement:

Be ready for multiple channels. People want to contact chatbots via their favorite messaging platform, such as WeChat, WhatsApp, Slack, or Facebook Messenger, or with voice assistants such as those powered by Siri. Platform popularity differs across countries, age groups, and use cases, so you can’t pick one and expect users to come to you. You need to be able to build a bot once that can run on many channels and can quickly add new channels, without having to rewrite it.

Have real conversations. It’s not that you’re trying to trick people into thinking they’re talking with a person (bad idea), but companies must tap into natural-language processing capabilities so a bot can understand, remember earlier comments in context, and respond in a normal, conversational way. Having real dialogues is as critical to a bot’s success as having a good user interface is to a mobile app’s success.

Use machine learning. A bot needs to get smarter as it works. That means it needs machine-learning algorithms to let it continuously figure out the many different ways customers ask questions. Think of how many ways we can ask, “What’s my checking account balance?” Tuning a bot to understand customer intent isn’t easy, so a bot-building platform needs these machine-learning tools built in.

Integrate to get the right answer. A chatbot’s intelligence comes from the data that it accesses—in back-end systems for data on customers, supply chain, production, marketing, and more. Bots need flexible integration that can tap into many sources.

Leverage mobile “moments” for great UX. Chatbots are great for a conversational UI engagement where the bot responds to unstructured end user questions. However, there are instances where users need to review a lot of data or enter a lot of data, such as filling out a loan application form or filling out a return form for a retail bot. These processes are not a great fit for a conversational UI. This is where the bot design should incorporate mobile moments. In mobile moments, the specific UI related to the prepopulated loan form or prepopulated retail form is surfaced automatically by the bot in the context of the bot conversation, rather than asking the user to go to another website or download a completely separate mobile app.

Know when to hand off to humans. A bot should be smart enough to know when it’s in over its head and needs to hand off to a human agent for the best quality of service. And human agents also should be able to hand off conversations to the bot when it can more effectively handle a request.

Action Items

How to Train Your Chatbot

What Does it Take to Train a Chatbot?

Watch this Webinar to explore the future of customer engagements with AI, chatbots, and mobile apps.

Photography by Fleur Treurniet, Unsplash