Cornell Tech Product Studio - Waiter AI
Introduction
Ordering food from a menu shouldn’t feel like a chore, it should be an empowered experience. Currently there’s decision fatigue from menus that can be overwhelming.
Waiter AI makes decision making easier, smarter, and faster with AI.
The Problem.
Ordering food when sitting in a restaurant is an unassuming problem. Its a low-stakes intensity problem. Most people feel like it’s too much work, and too much uncertainty, and they settle & take more risk than they want to in this environment.
Though the stakes are relatively low, the reality is that its a huge opportunity in terms of volume - just imaging the people ordering at restaurants every day across the US, it’s a huge repeat behavior.
We gathering data through customer surveys, and found that the average score of ordering confidence was only 3.5/5 across users. Only 11% people said 5, with most people scoring between a 3 and 4, which aligns with the analysis.
And when you empathize and try to understand where these numbers are coming from a few issues come to mind. Menus are an information overload, it’s a complex process of reading, imagining, and picking & choosing what you give your attention to.
The waiters are there to help you guide your ordering process. However, those interactions are also hard. Their recommendations can seem impersonal to what you’re looking for and biased, or no matter how they describe the food to you, it still might not click.
That’s where we saw the opportunity to build Waiter AI - tackle a hard problem that is small in intensity but big in quantity, and use AI to facilitate this interaction between customers, waiters, and static text.
The Idea: What is Waiter AI?
So we created a digital waiter assistant that acts as a Waiter and helps you with what you want to order.
A smart assistant that acts like a digital waiter:
→ “Craving something spicy and under $15? Got it.”Explain how it guides users interactively through mood, craving, budget, or dietary needs.
Emphasize conversational UX + fast suggestions.
How it works:
Take a picture of the menu -> OCR to create a data base of food items from the menu. We used this as input since it’s quicker and more reliable than building our own data base.
Generate the data - we augment the input data with more data such as flavor profiles, descriptions, why we think you might like it, and pictures, to help the user have a full experience of what they might expect
Of course, since the pictures are generated it may not be accurate, but we found that most dishes are fairly standardized.
What’re you craving today? - ask the user for input to create recommendations
Ex. I’m looking for something that is not too spicy with seafood
Recommend and interact - if you have further questions or want more recommendations, simply ask.
The Impact.
We conducted an ethnographic study in a real restaurant environment, and tested 3 products against each other - 1) the traditional menu & waiter, 2) competitor apps (yelp, google) 3) WaiterAI
By going through exercises of ordering, we were able to show 2 key impacts:
There are 3 stories in this data chart. 1) Static menus start strong, but diminish over time with no room for improvement. Consumers either know what they want, or have trouble finding it. 2) Competing apps take too long and don’t have a big impact -> long times, low increase in confidence. 3) our app shows improvement over time.
Our app showed lower time spent and higher confidence compared to static and competitors.
Important takeaways:
AI isn’t just a tool for efficiency, it’s a new way of interacting.
Compared to human waiters, we saw more interactions and questions.
People don’t mind asking robots annoying questions to get answers, and don’t worry about the judgement of looking uninformed.
AI, using ML, can efficiently understand data and recommend items to users - this is a big and broad implication
Generative AI is great for creating augmenting data to include imaginative descriptions, images etc.
Closing Thoughts, and Path Forward.
It doesn’t stop here - AI will have huge implications on how consumers make decisions and interactions going forward. A powerful tool, it should be studied not just for it’s efficiencies, but also the interpersonal medium it provides.
If you have any questions or want to see a demo, feel free to contact me.
Going forward, I’ve pivoted from the restaurant industry to retail/e-commerce more broadly, and currently building out a pilot product as LayeredAI.
