At San Carlos, Calif.-based NatureBox, a startup that delivers to users a monthly box of healthy snacks, data is playing a key role helping people find new nibbles to munch on–and keep them coming back for more. Over the last few months, the company has developed an algorithm to better recommend snacks that are in line with users’ individual tastes. On Wednesday, NatureBox debuted a new feature that helps subscribers compile their snack boxes while feeding data into the company’s recommendation engine. Customers can prioritize up to 20 items in their snack pantry, and the service will automatically send the user the top five items each month. To appeal to snackers who don’t want to commit to a subscription, the company also plans to launch NatureBox Market, an a la carte option it is currently testing, in the summer.
“I think we’re just starting to scratch the surface of what’s possible,” CEO and cofounder Gautam Gupta told Fast Company. “[The snack pantry] is giving us an indication of who the customer is and what products they like. From there, we’re able to improve our snack recommendation engine.”
NatureBox offers a line up of 120 to 130 snack options, and receives 10,000 to 20,000 new ratings every day. On top of its hundreds of thousands of ratings, the company also has information on customers’ dietary restrictions, allergies, and preferences for different flavors, thanks to a quiz new users are asked to fill out. On a most basic level, the algorithm can filter out snacks with allergens or non-vegan ingredients if users indicate they have allergies or follow a vegan diet. But the aim is to delight customers looking to try something new–users can opt for a “surprise snack” each month–or substitute in a similar snack in the event the one requested is seasonal or not in stock.
“We have so much data, and we’re constantly improving the performance of the algorithm,” Gupta said. NatureBox, which in July plans to open up a 40,000-square-foot facility in Indiana to serve the Midwest and East Coast, could incorporate other types of data in the future, such as the location of a user, time of day when he or she logs in, popular combinations of snacks, and more, he added.
“We’re starting to use other signals that might be seemingly unrelated to snack ratings or how you might enjoy a particular snack, but as the data set gets larger, we can start to correlate those things,” he said. “All of this is driven through our direct relationship with the customer. We really have a data set that traditional brands that distribute to retail don’t because we know who our customer is.”