As CPTO, John spearheads the advancement of kea’s Voice AI. In this interview, we dive into the opportunities and challenges of this increasingly popular Voice technology.

How would you describe kea’s product? How does it work? 

There are three fundamental functions required of kea’s Voice AI which are 1) hearing what the guest says 2) understanding what the guest says and 3) speaking back to the guest. For a Voice AI to hear, we need Automated Speech Recognition or ASR, for it to understand, we need Natural Language Understanding or NLU, and for it to speak we need text-to-speech or TTS. 

However, given that our Voice AI specifically serves fast casual and quick service restaurants, we couldn’t use just any ASR, NLU, or TTS model. We needed to develop unique models that could handle the complexity of our dataset – menus. Restaurants have hundreds of menu items and potential combinations, most of which have names and pronunciations that don’t exist anywhere else in the world, except for in that restaurant chain – and no restaurant brand is exactly alike. 

So, kea’s product works by using a restaurant’s menu to drive our unique ASR, NLU, and TTS models, which learn from guests as they place orders. When deployed, kea takes guests’ orders, confirms them, and speaks to the guests as if it were a cashier at the restaurant they’re ordering from – that’s why we are known as the “Cashier in the Cloud”. 

What are some of the biggest challenges you’ve encountered in developing kea’s product, and how have you overcome them?

To understand the challenges, you’d first have to consider that our product is different from most technology products today because there’s no screen or visual user interface. Unlike a chatbot or ChatGPT, kea delivers its user experience through audio only, offering a dynamic voice experience. The conversation is the product and, when ordering food, the conversation flow is crucial. The end goal of every conversation is a transaction, which ultimately affects a restaurant’s ROI. If our Voice AI says the wrong menu or the next step in the order sequence doesn’t make sense, the guest will most likely hang up. 

We’ve overcome this challenge by implementing an intuitive order flow and a tiered approach to handling orders, making our guest experience superior to other Voice AI products. We take as many orders as we can through AI automation, but if a guest is stuck or struggling, then a customer service representative will take over in real-time, without the guest needing to repeat their order. If the call drops, the representative calls them back. And if the representative can’t handle the call, the guest is finally sent to the restaurant’s manager. With this approach, we never give up on a guest, and as a result, have achieved a high conversion rate for our restaurant brands. 

How do you ensure that kea’s voice AI integrates seamlessly with existing platforms and technologies?

We’ve architected kea so that all of the platforms and interfaces we leverage can be easily reimplemented in other technical environments. While kea isn’t currently working with all Point-Of-Sale (POS) systems, we have the potential to be. Given that the POS system has access to the restaurant’s menu and has order-taking and payment-processing capabilities, we can connect to that POS. Currently, we’re partnered with Olo and CBS Northstar, leaders in the restaurant technology space, and have more integrations in the pipeline. 

In your opinion, what makes kea’s technology exciting to work on? 

We’re in an exciting time in this industry, where we have these amazing ML and AI tools, impressive compute power, and model architectures that are accessible enough so that even a relatively small team can make massive strides. Large Language Models (LLMs) for example are rapidly becoming more capable, faster, and cheaper, and they’re an excellent tool to leverage for building interactive conversational interfaces, but they’re only a tool. 

What makes kea particularly interesting is that we’re more than a tool. We offer a service that restaurants are in desperate need of, given the increasing labor shortages and operation costs. When kea answers the phone, we become a part of the restaurant’s brand and part of the restaurant’s revenue stream. 

Where would you like to see kea within 2-3 years? 

Within a few years, we want our guest experience to be equivalent to the absolute best store employee you can imagine. By that time, guests will prefer to speak with kea’s Voice AI because it’s so easy, efficient, and consistent. It’s exhilarating to be leading the team that’s making this ideal a reality.