March 1, 2018

The 8 Key Components for Successful Conversational AI

Parts of a disassembled typewriter in organized columns and rows.

There’s a lot of buzz about conversational AI, and with good reason. As human beings, we feel at ease in natural language conversations. Now that machines are getting better at understanding what we say and are responding to us in our own language, the door has opened for a much more effective way to engage with customers and employees.  

That said, building an experience that’s delightful — and not just annoying — means more than just spinning up a simple chatbot. Instead, you need to give careful consideration to each of the eight key components of a conversational AI application. In this post, I’ll introduce each component and show you how to to ensure the success of your conversational AI strategy.

We’ll cover both types of conversational AI applications:

  • Conversational agents, which allow you to have a conversation with an automated system in natural language via text or voice.
  • Conversational enhancements, which use AI to improve the quality or efficiency of conversations between humans.

Some of these applications can suggest a response to an email, learn who your company’s experts on a particular topic are, or automatically intervene in online bullying.

The eight key components to conversational AI are:

1. A Clearly Defined Value Proposition for the User

You’ll first want to consider how your customers and employees currently interact with your business and products through web, mobile, social media, email and other channels. If the interaction is already conversational, think of cases where friction or inconvenience exist. For instance, where people may need to wait a while for an answer, where they need to switch between apps, or where they need to input information repeatedly. If the conversations are primarily informational in nature, they may be good candidates for automation or partial automation through conversational AI. If they are more complicated, they may be good candidates for conversational enhancement.

Apart from existing conversational interactions, it’s also worth considering where conversations currently aren’t scalable but could be engaging and valuable. For example, you might love to have a personalized conversation with each of your users to introduce new product capabilities instead of sending a mass newsletter or developing a static FAQ.

2. Design

For conversational enhancements, this includes developing an engaging persona, choosing the right messaging platform and channel, perfecting the dialogue flow, and making sure a conversational interface is well-suited to the task at hand (see Can Your Chatbot Pass These 7 Tests? for more information). For conversational enhancements, you’ll want to identify the right points in a conversation for the system to offer suggestions to the human agents or users, and design the interactions so that they are seamless and natural without being intrusive.

3. Content

Since both conversational agents and conversational enhancements will enable communication with your users, you will need to decide how to develop the content that they share. If you already have conversational data available, you can curate the best of it and use that as the basis for the responses that your conversational AI application provides. Where you’re missing conversational data, you’ll need to use human writers or natural language generation techniques to fill the gaps.

4. Data

A successful conversational AI experience relies on training data from similar conversations as well as contextual data about each user that can be used to personalize the experience. For instance, you might use demographic data, user interests and preferences, or transaction history to inform what, when and how you communicate.

5. Language Technology

If you’re working with voice interfaces, you’ll likely need to use speech-to-text transcription to generate text from a user’s input and text-to-speech to turn your responses back into audio. For both voice and text interfaces, language understanding techniques such as sentiment analysis, question classification, intent recognition, and entity and topic extraction are likely to be relevant to understanding what the user is saying.

6. Other Machine Learning Capabilities

Besides language technologies, you may want to use machine learning models that help set the stage for a successful interaction and add value for the user. For instance, you may want to predict the optimal time and place (i.e., channel) to start a conversation with a user or automatically make product recommendations based on user history.

7. Feedback Loops

Each conversation should update your understanding of the user as well as improve your ability to craft a successful conversation. You might ask the user for feedback directly at the end of the conversational experience or look at downstream behavior (for instance, whether they re-engage or if the conversation leads to conversion), and use this knowledge to improve the next conversation.

8. Privacy and Security

Because a conversational interaction can feel so casual and natural, conversational data may often contain sensitive information that requires careful handling from both a technical and policy perspective. At the same time, you’ll want to be sure you’re able to make use of the data you’re collecting to improve the user experience down the road.

Some Final Thoughts on Conversational AI Components

When all of the components above are in place, a conversational AI experience can tap into many of the aspects that make human language such a versatile and rich communication medium. Your value proposition and content make the conversation worth having in the first place. Language technology provides the translation layer between your application’s logic and the user’s intent. Great conversational design respects the unspoken rules that people are used to making in conversations. That way, they’re not distracted by poor dialogue flow or badly handled failure cases.

Data, feedback loops and machine learning techniques give your application skills and context, much like a human conversation partner accumulates as they speak to you. Privacy and security enable the interpersonal trust often fostered by human-to-human conversations. Together, they’ll enable you to deliver an experience that really takes advantage of the potential of conversational AI and solves user needs in a natural and intuitive way.

This is my take on conversational AI components.  I’d love to hear yours. Connect with me on Twitter (I’m @jbrenier).

Growth insights
in your inbox

Join our community of thousands of tech entrepreneurs to get actionable insights from our monthly newsletter.