Dissecting the 8 layers of Conversational AI

Étienne Merineau
April 5, 2019

You’ve been reading it in major business and tech publications across the globe. It’s popping up in strategic meetings in boardrooms of Fortune 500 companies… Conversational AI and chatbots are on everyone’s lips. This relatively new term refers to “the use of messaging apps, speech-based assistants and chatbots to automate communication and create personalized customer experiences at scale,” and it is anything but a fad.

In fact, it’s a complete game-changer that will impact every aspect of customer experience, from product to marketing to customer service. According to Gartner, 25% of customer service operations will use virtual customer assistants by 2020. But before going all-in on the next big thing, it’s important to understand the many moving parts involved in creating a world-class, enterprise-grade Conversational AI assistant. At Heyday, we know a thing or two about what it takes, so we’ve created this handy guide to help you dissect and understand the 8 critical layers of Conversational AI.

Chatbot, /ˈCHatˌbät/: A computer program or artificial intelligence designed to simulate a conversation with human users.

1) The Decision Tree: Mapping the possibilities

Before walking around a new city, you need to know where you’re going. It’s no fun thinking you’re going for a nice stroll, ending up in the industrial district, and then even worse, not knowing how to get back to the hotel. The same goes for a chatbot. The user needs to be guided in some way because as soon as they choose to interact with your conversational assistant, they’re going in blind. The decision tree is essentially the map of your entire chatbot and serves as the skeleton of your user experience. It outlines the different paths (or “flows”) your user can go down.

For example, when a user first interacts with D4, Decathlon’s Conversational AI assistant powered by Heyday, it provides users with a selection of options to get started. They can start by searching for a product, checking out answers to frequently asked questions, or learning about Decathlon’s membership program.

Each of these flows is subsequently scripted and guided, but it’s up to the user to make the decision to move the experience forward, and it’s the responsibility of the conversation designers and programmers to make sure that no matter what the user chooses, the chatbot has a proper response and appropriate next steps for them. A little bit of hand-holding can go a long way in creating a delightful experience.

2) Company-specific AI

AI is where the tricky part (and the fun) begins; it is the brain behind your chatbot. Through the use of Natural Language Processing (NLP), the AI makes sense of users’ intents and responds to questions based on its understanding and context. However, your chatbot’s AI is only as good as you make it. Because while the decision tree offers a roadmap for the user, sometimes the user goes off the beaten path, and your chatbot needs to be prepared for it. In order to accurately understand what users ask, your AI must be trained on a variety of content and possible sentence and question structures.

Company-specific AI essentially covers all the questions, policies, and information specific to your brand. This includes but isn’t limited to: return and refund policies, shipping and delivery policies, opening hours, store location(s), membership and loyalty programs, contact information, and more.

That means that if users don’t have time to click through the chatbot experience to see what your opening hours are, your chatbot should be able to understand various formations of each of these questions, like “When are you open today?” “What are your hours?” “Are you open on Tuesday?”, etc. Over time, this critical layer will help build the brand’s brain and ultimately help turn your brand into a helpful, resourceful customer sidekick.

3) Product-specific AI

So you’ve solidified your decision tree and you’ve laid down the foundation of your company-specific AI. Now it’s time for product-specific AI. This accounts for everything related to your brand’s field, whether it’s sporting goods, vacation booking, hardware and home improvement… Product-specific AI covers everything about what you’re selling. If you’re a sporting goods brand, your AI should be prepared to answer everything from “How much flex do your hockey sticks have?” to “Show me the best hiking shoes.”

At Heyday.ai, we do this by integrating with our clients’ online store’s product feeds. Whether it’s Shopify, Magento, Lightspeed, Salesforce, BigCommerce, PrestaShop or another ecommerce platform, our AI assistant launches queries and scrapes the information from the database, giving customers responses to specific products in real-time.

If the question is too complex or personal (edge cases), our AI assistant will escalate it to a product expert in-store where it gets into “listening mode”. It then monitors the conversation to enrich its knowledge base with product-specific information. Over time, these interactions between the AI assistant and the brand’s sales and support teams create a virtuous circle that help foster a richer, more resourceful AI.

A mobile screenshot of conversational AI featuring Decathlon Brossard, along with a product carousel.

4) Industry-specific AI

Next up is industry-specific AI. These are questions that are particular not necessarily to your brand, but to your vertical. For example, if you’re a vacation booking provider, your AI will need to be trained on common inquiries such as, “I need to change my booking.” “Where is my hotel?” “Can I upgrade to first class?” For apparel brands, this means being able to answer “What fabric are your pants made of?” and “Can I wash this in warm water?”

This layer of AI is crucial especially if you want to stay competitive with other brands in your space, since other vacation booking brands, apparel brands, etc. will be training their AI on these same inquiries as well, so it pays to take the time to make it particularly robust.

Industry-specific AI must also understand inquiries based on the context of your brand’s product. So if a user asks a snowboard brand’s AI about a superpipe, it will know that the user is probably not looking for something to fix their sink.

Screenshots of Activia conversational AI assistant interacting with a customer allergic to gluten.

5) Contextual AI

To foster long-term customer relationships, your chatbot should be anything but ephemeral, but you need to take into account those important days of the year that come and go. For example, if you’re in retail, ecommerce, or you ship consumer goods, you better be prepared for those holiday season questions: Are you open on December 24? What are you Black Friday deals? What time do you open on Boxing Day? What’s your delivery policy on Christmas? Maybe your customers want to know if you have any special sales for Mother’s Day, or when you’ll get your winter stock in.

Whatever the case, you need to be ready for those temporary, fleeting opportunities to build lasting customer relationships.

Seasonal and contextual insights help brands leverage specific moments of the year to adapt their pitch and stay relevant. At the end of the day, customers expect a quick and accurate answer to their questions, no matter the context. It’s up to brands to build a resourceful AI that is aware of the temporality of the conversation and can adjust on the fly.

6) Linguistic AI

It’s important to create a chatbot that’s not only accessible and accommodating to as many customers as possible but communicates your authenticity and humanity. In order to remain culturally relevant, this means that you might have to create a multilingual AI, or at the very least incorporate local slang and take into account specific customs and proper manners. For example, in Canada, you’ll have a large chunk of the population speaking either English, French or a homegrown mix of both i.e. Frenglish. So that means that your AI needs to be prepared for a whole lot of “Do you speak English?” and “Je ne comprends pas l’anglais”.

On another level, there’s a huge difference between Parisian French and French-Canadian French. For example, when a Parisian refers to shopping, they will say “faire du shopping”, whereas a French-speaking Canadian will say “faire du magasinage”. Making sure your bot uses the right expressions for their language and their region is fundamental to speaking to customers in a way they are familiar with and creating a personalized and relatable experience where the customer feels understood.

7) Universal AI/Small talk

Generally, people don’t mind talking to bots as long as brands set the expectation that users will be, in fact, talking to a bot. Nothing is more frustrating than expecting a human answer and getting anything but (and not understanding why.) But when people know they’re talking with a machine, one of their first instincts is to test it with weird and funny questions. Just ask Siri.

Which brings us to the fun stuff, universal small talk. This might be some of the most important because a bot’s ability to respond to small talk is how people will rate just how smart your bot is. “How’s the weather?” “How are you doing?” “Are you a robot?” “Will you marry me?” These are just a few of the questions that people might ask not because they care what a machine thinks about the weather necessarily, but just to see how much your robot can really handle, and how human it can really be.

This is where you’ll find out if your AI is really robust enough to deal with all the random, seemingly mundane questions people have for it. They’re not testing your brand, they’re testing your bot’s intelligence. But like any other layer of Conversational AI, it’s only as good as you want it to be, and training it and preparing it for human unpredictability is crucial. Full disclosure: it is a long term process, since you first need to go live and listen to conversations before you can learn from them and ultimately enrich your AI progressively over time.

A mobile screenshot in french where the bot is engaging in small talk.

8) Personalized AI

Up until now, you’ve been training your brand’s AI yourself, but now it’s time to take off the training wheels. Personalized AI allows your conversational assistant to learn about your customers’ based on past interactions in order to augment and personalize each individual’s experience.

This can mean proposing product recommendations, reminding customers of reservations they made, tracking orders and deliveries, reminding customers of abandoned shopping carts, and more. While the other layers of AI are more reactive (customer asks questions, AI provides the answer), the AI becomes more proactive and anticipatory in this personalized level.

The goal is to re-engage with customers based on their conversation history, preferences, habits and intents. Essentially, this is the holy grail of marketing and commerce: where every experience is bespoke and adapts to the shopper based on real-time data and context. Only AI can help brands create this level of clienteling at scale and treat every person as a VIP, without making operational costs explode.

Ever-learning. Ever-improving.

At Heyday, we get a front row seat to exactly how users interact with conversational assistants. We’re always making sure to regularly comb through our logs to see what’s working, see what’s not, and subsequently train our AI accordingly to meet the specific needs of each client. That being said, each of these layers of Conversational AI is crucial to any chatbot worth its salt, but our job is by no means done just by simply checking these boxes. Once you’ve established each of these layers of Conversational AI, you’re off to a great start, but it’s just that, a start.

AI assistants are very different by design compared to other marketing tools like websites or mobile apps. Day one (or launch day if you prefer) is the worst time to judge a brand’s AI assistant since it’s still in its infancy. To gain that crucial data, it needs to interact with customers to expand its knowledge base and its capabilities.

Since it’s iterative in nature, Conversational AI is a long term investment, and there are always improvements to be made to make AI more intuitive and customer experiences more seamless. The possibilities are as endless as you want them to be, but a little bit of patience will be the key to unlocking the true potential of this new customer communication channel.

Opening up the floodgates of instant messaging means you’ll be handling the unpredictable nature of human beings and each customer’s way of writing, speaking and chatting. To craft an AI that can understand and handle all of these might seem like a daunting task, but it is entirely within your grasp. Plus, conversational AI is much more than a trend; it’s an opportunity to gain a competitive edge for the future. It’s about laying the foundation for better, more helpful brands and more meaningful customer relationships.