Sharpening the leading edge of conversational AI on a daily basis, Jordan Prince Tremblay is Heyday’s director of data and AI. Here at Heyday since the very start, he created our approach to AI from the ground up— also helping to build a great team at the forefront of development in the industry. No one knows the ins and outs of the retail sector’s top conversational AI platform more than him or has quite the same insights about the market as a whole. 

To illustrate how advanced conversational AI is getting, 27% of consumers in a PricewaterhouseCoopers survey said they’re not sure if their last interaction with customer service was with a human or bot. Heyday conversational AIs take the extra step of always introducing themselves to customers as virtual assistants—and the product in more ways than one speaks for itself. The data too. More than 50% of consumers choose AI over humans to save time: a key benefit of conversational AI—one of many.

Tremblay shares his experiences with the technology in this exclusive Q&A, debunking some popular myths about AI and predicting what’s coming in 2022. A little preview: conversational AI will take a huge leap forward. And brands and their customers will reap the rewards. The Heyday platform will keep evolving too. Tremblay has all the details:

Myth: Automation and AI are basically the same.

Jordan’s take: There are several big myths surrounding conversational AI. First and foremost: that it and automation are one and the same. While AI does automate processes, it goes beyond that. 

Instead of systematically repeating the same task, like on a production line manufacturing cars, AI mimics human intelligence to make decisions or even create new content. It’s not just repeating the same process, but also the ability to distinguish between multiple choices, make a decision, and then gain insights.

Myth: Conversational AI is simpler than other forms of AI. 

Jordan’s take: Conversational AI is a subset of AI, which is a really generic, broad term. You can use AI to solve problems or find the best path to go somewhere, but you can also use AI to understand humans. In conversational AI, computers have to first understand before they can take action during a conversation, but conversations are complex interactions. There’s a lot of nuance and the sequence of the conversation has a big impact on the meaning. Sentences that are 100% the same except for one word can have different meanings altogether. But there are also different ways to ask the same question, like: “When are you open?” or “What are your business hours?”

Languages are already really complex. There is local jargon, business-related vocabulary, and acronyms. When interacting with conversational AI, people use slang, make spelling mistakes, switch languages, and change topics, making the process even more complicated.

Myth: Conversational AI is just a fancier term for chatbot.

Jordan’s take: What we used to know as simple chatbots have really come a long way. Natural language processing (NLP), understanding (NLU), and generation (NLG) are the three pillars of AI-powered conversational AI chatbots—or virtual assistants—today. They separate conversational AI from older chatbots thanks to contextual understanding and personalization. 

Early bots, like Dr. Sbaitso (Sound Blaster Acting Intelligent Text to Speech Operator) in 1991, gave customers purely scripted experiences. Natural language wasn’t an option for developers and people could only have simple “conversations” with Dr. Sbaitso, which acted like a psychologist. Users got limited responses back, including “That’s not my problem,” when the doctor was asked a question he couldn’t understand. It’s like moving around in a maze, where you can only go in one of two directions, but everywhere else is blocked. It’s constricting to say the least. The industry is understandably moving away from that setup in favor of more engaging CX.

NLP simplifies the task of parsing through text for the meat of a customer question—removing filler words. That helps the AI understand the nature of the question. And NLU, which interprets customer intent, is a subset of NLP. NLU mimics the cognition of humans to understand what is the meaning of a sentence, what is being said. The AI needs to understand whether you’re looking for the closest store or looking for a deal. 

When the chatbot is ready to reply, we get into NLG. That’s where the AI fetches information from external sources, like products available in inventory, and then creates the dialog saying, “here’s the best deal I’ve got for you.”

Myth: Conversational AI innovation will probably plateau.

Jordan’s take: Based on how far the industry’s come in a few short years, there are no signs of it slowing down. For starters, we have many tools now that just weren’t available then, tools used to standardize conversational AI, to manage datasets, create models more easily, etc. 

There is a lot of effort in the community to bootstrap conversational AI from scratch, because you may not have prior data to draw from, but it’s not only about monitoring. It’s also about making sure the entire system stays relevant. The ability to iterate faster on quality AI models— letting it adapt—has accelerated adoption with retailers. 

After all, people ask different questions in the summer than they do in the winter, because they’re buying bikes instead of skis. Meanwhile, the ability to integrate different channels, like Facebook Messenger, WhatsApp, and Google’s Business Messages, has definitely been a game-changer: new channels means more places brands can connect with customers, which has been really important, especially over the last few years.

Myth: All Conversational AIs are created equal. 

Jordan’s take: The quality of conversational AI can change drastically from chatbot to chatbot. It all depends on the work being done on the back end. Personalization is just one aspect of conversational AI where Heyday stands out in the market, customizing the tone of each brand’s conversational AI—and the customer journey of each shopper. NLP, NLU, and NLG may be the three pillars of conversational AI, but cost and quality join personalization as Heyday’s three biggest pillars.

The AI has to bring return on investment (ROI). So, we make sure our solutions are cost-effective. Then, getting high-quality AI means going to the fine-grain level and personalizing as much as possible. It’s a big challenge. The more precise you want to be, the more resources you need, because the development process becomes more complex. 

We use different types of algorithms to solve the understanding part as much as possible, to personalize the customer journey for each client, but also really high-quality algorithms for recurring themes and functions in e-commerce, like tracking orders and searching for products. That’s how we balance these pillars.