Tech company Crisp forecasts demand with machine learning, reducing seafood waste
It's one of the questions that bedevils companies at every step in the seafood supply chain: How much demand exists for your product, right now, and how much do you need to ship?
Many demand forecasters rely on trusty sales spreadsheets packed with historical data – plus gut instinct informed by years of experience. But a new company seeks to take the guesswork out of demand forecasting by replacing spreadsheets with machine learning and gut instincts with data.
Crisp launched about two years ago with the aim of using technology to reduce food waste across the food sector at all stages of the supply chain and has so far raised about USD 14.2 million (EUR 13.1 million) in funding. The company released its alpha product to 30 companies about five months ago, and is now offering it to the general public. More than 100 customers are currently on the platform, with five or six new ones onboarding every week. Several seafood companies, including Norwegian salmon company Hofseth, have joined.
Customer data is uploaded and fed into a machine learning algorithm that combines that data with information about external factors such as holidays and seasonal trends to generate a forecast. Crisp Founder Are Traasdahl was inspired to work on combat food waste after a family trip around the world in which he witnessed its staggering scale.
"We were in New Zealand and saw apples rotting in the field and we were in India and we saw kids my kids' age starving in the streets," Traasdahl told SeafoodSource. "As technologists, we were curious about solving that problem."
Predicting demand – and matching supply to that level – is a complicated task influenced by many factors that older technology systems are not equipped to handle. Seafood, with its short shelf-life and convoluted supply chains, depends on accurate demand forecasting.
"Seafood is an incredibly complex supply chain," Traasdahl said. "It's a fresh product. The fresher the product, the more need there is for forecasting. There is a lot of logistical cost for seafood because it's generally not fished in the place it's consumed, so there's lots of logistics and storage costs."
"If you produce too much, you have to throw it out and that's expensive,” Traasdahl added. “And if you produce too little and end up short-shipping, you lose revenue and you lose trust of customers if you don't make shipments.”
Traasdahl has worked in technology for 20 years, and has launched three companies. When starting those businesses, he faced hundreds of competitors. This time, his main competition is either large-scale companies such as Oracle or SAP – or Excel spreadsheets.
"Ninety-nine percent of the companies out there use Excel or some kind of technology like Excel to do forecasting. And it's usually one very smart person with one very large Excel spreadsheet and they take all the factors that are influencing demand into that sheet," Traasdahl said.
Before onboarding with Crisp, most customers don't have much ability to create sophisticated forecasts. Instead, they might look at the previous year's sales and add a nominal percentage on top of that. According to Crisp, customers who used to achieve 50 or 60 percent accuracy now get 90 or 95 percent accuracy. In addition, they get an 80 percent reduction in food waste and a 10 percent increase in revenue.
At this point, Crisp keeps each customer's data siloed from other customers, though the company could have the ability to generate information on macro-level trends by combining customers’ data. Crisp plans to incorporate that type of macro data into forecasts "pretty soon," Traasdahl said.
At Traasdahl’s previous companies, customers found it useful to compare their performance to industry benchmarks created using anonymized data from many customers.
"We haven't launched that yet," Traasdahl said. "Customers find it very valuable to have those indexes and also to see what happened in categories outside their own categories, so we definitely have that part of what we're discussing."
So far, companies with revenue between USD 50 million (EUR 46 million) to USD 5 billion (EUR 4.6 billion) have adopted Crisp’s product the quickest – especially food brands in that size range. But retailers with USD 10 billion (EUR 9.2 billion) in revenue are also joining the platform.
Multiple pricing tiers – from free to USD 5,000 (EUR 4,600) per month – offer solutions for companies of any size. Depending on the depth and volume of a customer’s data, Crisp has a handful of different algorithms it can use to generate forecasts.
"We have the ability to use a specific type of algorithm that's best suited for a particular type of data set. If we have a different data set that is deep for many years, we have a different algorithm for that," Crisp Vice President of Products Trevor Hough told SeafoodSource.
Customers can even retroactively add events to the data to help the algorithm better predict future events, while relying on past data sets to create forecasts of brand new products.
In the end, the platform significantly cuts down on the amount of time that demand forecasters spend trying to create accurate estimates.
"They go from spending 15 to 20 hours per week – each person on the planning team spends that – and they go down to one hour," Hough said.
Photo courtesy of Spring Capital Partners