Continue to Site

    The Myth of Artificial Intelligence in Restaurants

  • AI still has a long way to go before it becomes truly intelligent.

    Domino's
    Domino's has tested AI in several functions, including for phone orders.

    Artificial intelligence is one of the biggest buzzwords in today’s markets and hospitality is no exception. I’m uncomfortable with the term.

    Almost every reference to AI in the restaurant sector today is actually an example of machine learning, and I believe there is a huge distinction between the two. AI still has a long way to go before it becomes truly intelligent. Machine learning is the first step, where behavioral data can be analyzed and used to improve a guest’s experience, but today it still requires a significant amount of human input to help it make decisions.

    For an example of the limitations of AI, let’s look at Pret a Manger, the international sandwich chain. They have a fantastic rewards program, which essentially empowers all their staff to provide a certain amount of free food and drinks to customers on a daily basis. Which customers receive these rewards is completely at the discretion of the staff member—the old man with the sad expression, the regular who always has a smile on their face, or the young woman who was just drenched by a bus driving through a puddle.

    The success of this kind of program is down to the human ability to identify or empathize with another person, based on a limitless number of potential factors. It works on both sides, giving the staff member a sense of wellbeing at being able to choose to make someone else’s day, and the recipient, who feels a similar sense of delight at being chosen … in addition to the free food, of course. That emotional recognition-reward loop is incredibly powerful. It’s part of what has driven intelligence for centuries.

    Currently, so-called AI technology wouldn’t be able to offer a comparable experience, and perhaps will never be able to. In addition to being able to rationalize why to offer someone a reward in the first place, the kind of intelligence needed to make this kind of decision would also require the AI itself to feel some sense of reward from the process. And if we ever get to that point, there will be many more ethical questions to ask that I won’t even begin to explore today.

    Baby steps

    A self-service kiosk could, of course, randomly offer rewards. But it would still lack the emotional intelligence needed to come close to the emotionally intelligent personal touch of Pret’s customer reward program. However, advances in machine learning are at least helping us get closer to the idea of artificial intelligence. Like learning to walk, we first have to take those awkward baby steps before moving onto more complex movements.

    If we again consider a self-service kiosk—with a little data integration, it can easily be “trained” to base its decision to provide rewards on the frequency of a customer’s visit or if they are a “high value” customer. Indeed, from a memory perspective, machines are already far superior to humans—they can accurately remember a customer’s entire lifetime of purchases, while a human server would maybe have some idea, but only in a fairly general sense. Anyone else get asked if they want “the usual?” when they visit their favorite coffee shop?

    If we go a step further, we could integrate the in-store kiosk system with local weather data, so it could choose to promote an ice cream on a sunny day, or a hot chocolate on a cold day. A step further still, machine vision could potentially allow the kiosk to recognize a guest’s mood or to observe the number of people in front of it and promote multi-person or family offers. OK, that last one has lots of potential for getting things wrong, but you get the idea.

    There is great scope to improve the way that technology harnesses all kinds of data to offer more relevant rewards and promotions, as well as better upselling and cross-selling… but that still isn’t intelligence in my book. For me, true AI in hospitality would mean technology that can evaluate, predict and take action in an “emotional” way. For example, this could include being able to accurately assess a situation involving a customer complaint and convert a potentially negative guest experience into a positive one. This kind of situation is a normal, everyday occurrence for serving staff, but is currently impossible for technology to do the same. 

    Mobile holds the key

    I believe that one way of getting close to something like AI in hospitality is via our smartphones and mobile devices. Today, our phones have become our closest and most valued possession. We spend more time with them than our real-life friends and family, and we trust their judgement on things like directions, weather predictions, and where to go and eat or drink. It is amazing to see just how much user experience professionals have tapped into the emotional engagement with machine and human. They play off basic human emotions, need for approval, anxiety (FOMO) and affirmation to create habitual experiences. Technology can only become intelligent when it can access enough data to make decisions based on more than just our purchasing trends, and our phones (and the cloud services they access) contain the most accurate digital reflection of our personality.

    Of course, accessing this kind of knowledge presents all sorts of privacy questions, but, from purely an AI perspective, if the information contained on a phone and its associated applications can be accessed, then the potential exists to dramatically increase the capability of machine learning into something much more akin to AI.

    Daniel Rodgers is founder of restaurant technology firm QikServe. Using any channel from kiosks and tablets to web and mobile apps, the QikServe platform delivers powerful solutions from ordering to payment, giving guests the convenience to order and pay for their food and drinks whenever and however they want. www.qikserve.com