TORONTO – For Krista Caldwell, the most shocking part of university was grocery shopping for the first time.
“I couldn’t believe that apples cost more than chocolate bars,” the cofounder of Vancouver-based Deepnify, which won the People’s Choice award at non-profit startup booster Next Canada’s NextAI Venture Day on Sept. 20, tells ITBusiness.ca.
“It’s shocking, because chocolate bars have ingredients from around the world – cocoa and nuts and caramel – and apples grow in Squamish,” she says. “But it turns out the reason is that 20 per cent of apples are wasted before anybody buys them in grocery stores, because the statistical forecasting tools that work really well for replenishing most items in the store fail on fresh foods with a short shelf life.”
That lesson led to the creation of Deepnify, a machine learning-based platform that Caldwell and her co-founder, Nima Shahbazi, have designed to help food companies predict daily demand for their products, and reduce foot waste, by analyzing a mix of historic sales data and consumer trends.
“It’s an important niche,” Shahbazi says. “If you lower the waste, you can lower the price, which is good for the customers. More customers is good for the stores, and it’s also environment friendly.”
It must be noted that as a startup, Deepnify is still developing its platform model, though it’s already built what Caldwell calls “the most accurate models in the world” for two international customers and is currently developing models for two Canadian customers. (We aren’t allowed to name them, but trust us – you’ve heard of them.)
The company’s secret weapon is Kaggle, a Google-owned platform that invites hundreds of thousands of data scientists from around the world to compete for and collaborate on projects – and Shahbazi is ranked number 19 internationally.
“Just like the best hockey players in the world compete in the NHL, the best data scientists compete on Kaggle,” Caldwell explains. “So we used some of those competitions to create proof-of-concept projects that predicted sales for two global food companies.”
A typical model, Caldwell says, will combine three years of historical sales data with statistics indicating customer behaviour based on weather, season, and other factors measured by Google trends to predict a given product’s sales, which are then forecast on a monthly or weekly basis – at numbers that are usually lower than their human-developed counterparts.
“So what we’re doing is building a team of the best data scientists in the world to tackle the expensive problem of food waste and create a zero-waste supply chain, so that eventually apples will cost less than chocolate bars,” she says.