VentureBeat presents: AI Unleashed - An exclusive executive event for enterprise data leaders. Network and learn with industry peers. Learn More
The recommender system uses a collaborative filtering, supervised learning model to match consumer preferences to foods. Customers answer questions about their dietary habits, the kinds of foods they (and family members) like, the family size, budget, and more. On a weekly basis, the Hungryroot algorithm predicts the groceries the customer might like. Once the customer approves the list, a box ships from one of three Hungryroot locations. Customers also receive a set of recipes, also predicted by the algorithm, that use the week’s ingredients.
Neil Saunders, the managing director of GlobalData’s retail division, has seen grocery retailers of all stripes lean into AI as a way of better forecasting demand. “With the disruption from the pandemic and more people buying groceries online, demand forecasting has become increasingly difficult for retailers and AI can help them make sense of the data and make more accurate decisions about what to stock,” Saunders says.
The AI-powered grocery challenge
Hungryroot works on a collaborative filtering model much like Netflix, learning from customer likes over time and pooling their preferences with others’. But AI-based recommendations for groceries are challenging, says CTO Dave Kong. For one thing, Netflix can recommend movies from a near-infinite queue. There are no additional constraints. Food, on the other hand, is not a consumable entity like movies. Food is perishable. Your choices depend upon inventory and on how much you can fit in the box.
An exclusive invite-only evening of insights and networking, designed for senior enterprise executives overseeing data stacks and strategies.
While consumers who like horror movies can feed on films in that genre for a while, the same need not apply to food. Feed consumers pasta three weeks in a row and they might complain. “The first step is to dissect the problem better for each customer. For example, what does variety mean to each customer? Is it different items (i.e. types of pasta like spaghetti vs. penne), or different dish types (i.e. pasta vs. salad vs. stir-fry, etc.),” Kong says.
Hungryroot is also trying to figure out what repetition means to the customer. “Are they looking for similar recipes and items to their last order, even if it’s two weeks ago, or does the skipped week matter to them? We can then focus on the right AI approaches depending on what we learn,” he says. “Understanding repetition and variety is the key to success in the food model that is not a factor anywhere else.”
The other challenge is that the number of customers who might like the exact same recipes using the exact same ingredients is not as large as movie buffs liking a genre. Consumer food preferences need to be digested at a much more granular level: salty, different types of protein, texture, and more.
The Hungryroot factorization machine crunches 60 different parameters (that number continues to increase) into its model. And data sources aren’t limited to only what customers say or do — Hungryroot also relies upon additional sources, like nutritional data.
A pleasant side dish: reducing waste
The Hungryroot algorithm optimizes recommendations not just for an individual user, but across the board for all its customers. Tweaking what’s in the box just a bit — if a customer likes one kind of white fish, they might like a similar one in large supply at Hungryroot — can help optimize food distribution across all boxes, cutting down waste, Kong says.
In addition, the AI-powered grocery suggestion algorithm itself is smart and helps Hungryroot to predict how much of each kind of food to buy. Since customer preferences are known, it’s easier to forecast demand and manage inventory. Saunders agrees. “The main advantage for brands is that they get better at providing customers what they want and have enough stock to satisfy demand. With regular grocery delivery, one of the most frustrating things is bad substitutions or unwanted products. If AI helps brands to understand what customers want they have a greater chance of building loyalty and repeat business,” Saunders says.
Hungryroot also makes sure to keep customers’ pantry purchases in mind: While every recipe might need salt, customers don’t need to buy salt every week.
Growing appetite for AI grocery delivery
Customers have responded well to Hungryroot: The startup is up 133% year-on-year for active customers. In June 2021, Hungryroot raised $40M in a series C funding round.
The algorithm has a high success rate. Consumers buy 72% of the AI-powered grocery deliveries. Kong expects including more unsupervised learning in addition to the supervised learning model. “We believe a neural-network model that is great at factoring in temporal information and excels at pattern recognition is the key to creating a successful and effective AI-enabled grocery service,” Kong says. “If we can nail the right level of predictability and variety for each and every customer, then we’ve solved the hardest problem with AI-enabled grocery shopping.”
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.