Katerina Axelsson is the founder and CEO of San Luis Obispo, Calif.-based Tastry, an AI-powered taste recommendation app used in grocery store wine departments.
Jon Springer: Welcome to the Beeak Room, Katerina. Tastry is a recommendation engine based on having “taught a computer how to taste.” Can you describe how it works?
Katerina Axelsson: Tastry is a SaaS (software as a service) and insights company, and a multidisciplinary innovator in machine learning, analytical chemistry and flavor chemistry. We created methods to predict how sensory-based products like wine, beverages, foods and fine fragrances will be perceived by consumers in general, and any specific individual in particular.
We like to say that we “taught a computer how to taste” because our core innovations are in our ability to break down the “flavor matrix” and how it relates to products and consumers. And we provide these insights to manufacturers, distributors, retailers and consumers.
What was the inspiration for its invention?
In early 2016, I was finishing my chemistry degree at Cal Poly, which is located in the central coast of California in wine country. To pay my way through college, I worked at winery and custom crush facility where we had many clients who wanted to make a wine label use our facilities, and we had winemakers to make and bottle their wine.
During this time, I noticed many idiosyncrasies within the wine industry. For example, you could have a 10,000-gallon tank of wine and sell half the blend to one customer and half to another. And the same wine could go under two different bottles, two different labels, and then sell for a different price and even receive a different score from the same critics. I saw an opportunity to pursue a more data-driven approach to winemaking that could benefit consumers, producers and ultimately the entire supply chain.
In free time, I ran some experiments, where I developed an approach and methodology to measure the compounds of sensory-based products like wine in a way that could be more closely attributed to consumer perception, and during the course of analyzing hundreds of wines, I had a series of breakthroughs and uncovered interesting insights that would benefit the field of flavor chemistry.
I took my results to a computer science PhD at Cal Poly, because I wanted learn more about creating recommender systems. I read a lot and wanted to learn more about alternative methods to collaborative filtering algorithms; methods where you don’t need a very large data set, or a lot of historical data, and a lot of users to derive insight. I had a unique and never before seen data set, and I wanted the best method to process it.
I scheduled a half-hour meeting this this professor... and this turned into a four-hour meeting. He brought several other PhDs into the room. They started arguing and drawing logic trees on the board, and long story short that professor has been our CTO for the past two-and-a-half years. Together we validated and are in the process of patenting our system and method for tracking and predicting consumer preferences for sensory-based products.
In your recent Groceryshop address, you mentioned that Tastry helped to bring category sales in wine up by 3% to 5%—and margins by 18%. How?
It’s all about going beyond personalization. We match the consumer preferences to specific products with a high degree of accuracy; we are 45% better at matching consumers to products they will like than the consumers themselves—or any other method they would use to buy a product or get a product recommendation. The odds were that the shoppers were buying the lower-margin wine that they’re typically exposed to, and we organically steered them to other brands that they wouldn’t have discovered on their own. We also saw a 35% increase in wine shopper satisfaction.
I’d imagine Tastry data has implications for manufacturers and brands. How are they being put into play?
We use our technology to determine how to maximize a product’s success before it hits the markets.
Aren’t some food purchases about things other than taste? An eye-catching package, a cool brand, a hot price? How, if at all, can a recommendation engine account for these subjective factors?
This is true for the first-time purchase. However, the customer will not return to the product if they didn’t like it. Tastry helps brands and retailers drive repeat visits and revenue, as well as customer satisfaction. As soon as a shopper buys into the Tastry recommender, they are more likely to try new products.
If AI works for wine recommendations, would it not also work elsewhere in the grocery store?
Tastry uses proprietary AI to automate food pairing to the wine based on the consumer’s preferences. We pair the wines in the store with over 1.5 million recipes that we recommend. We are also launching our beer recommender in late 2019.
Coolest places to shop for food around San Luis Obispo?
The local farmers market every Thursday night downtown.
What was your first job?
I worked at a yogurt and panini shop by the beach.
Burrito or burrito bowl?
Burrito bowl—more room for the good stuff.