AI could solve Japan’s food waste problem

Each year, Japan discards 6.4 million tons of food that is still edible, according to a government estimate. A Japanese weather services provider wants to cut down on that waste by unleashing the power of artificial intelligence.

The Japan Weather Association estimates that over 30% of industries are subject to weather-related risks, but the food sector is particularly susceptible. So it has partnered with retailers, food producers and other companies to develop a new system for predicting food demand. The system analyzes not only weather information but also sales data and other factors to project trends. This promises to help companies scale back excess production and reduce inventory losses.

About 30 companies are on board with the project. The research focuses on several dozen products, including cold ramen sauce from Mizkan Holdings, tofu from Sagamiya Foods, coffee from Nestle Japan and carbonated drinks produced by Pokka Sapporo Food & Beverage. Convenience store chain Lawson and other retailers provide sales data.

Last year, the JWA, Mizkan and other partners made some preliminary predictions using atmospheric temperature data from European organizations, which offer more precise information than forecasts by the Japan Meteorological Agency. The JWA reckons that by adjusting production according to such predictions, companies would reduce the amount of wasted noodle sauce by 40% and discarded tofu by 30%, for example.

In addition to monitoring sales and weather patterns, the artificial intelligence system looks at Twitter to get a read on shopping habits. The system is already capable of making predictions on, say, the products women in their 30s tend to buy on rainy days.

Into the Twitterverse

It is not unusual to use temperature, humidity and other weather data to forecast demand. But consumer sentiment is trickier to adjust for. A hot day can affect shopping patterns in different ways, depending on whether it occurs in the spring or summer.

Seasonal products — like cold noodle sauces — are particularly sensitive to consumer mood swings. This is why the JWA is tracking tweets. AI analysis of how frequently words like “cold” and “hot” appear can offer a deeper understanding of consumer behavior.

Better, timelier predictions could make for smoother distribution. Armed with a reliable weather forecast two weeks in advance, a manufacturer might opt to transport goods by sea rather than overland. This could cut costs and keep carbon dioxide emissions in check.

The JWA plans to increase the volume of data the AI system sifts through. The more data the system has on consumer habits, the more accurate its predictions should be.

Of course, food manufacturers, distributors and retailers make their own projections using point-of-sale and other data. But the quality of that data varies, it is rarely shared between industries and companies, and a lot of it simply goes unused.

The result is a fairly wide margin of error when setting production and order volumes. The process also requires personnel capable of understanding the data and making decisions based on their own experience. Shorter product cycles and sudden weather changes complicate matters.

But weather prediction and other data analysis technologies are improving. The JWA sees an opportunity to narrow the margin of error, share data intelligently and help businesses that lack the staff to deal with reams of information. It intends to offer the system to smaller companies starting in 2018.

Give it a share: