Canadian Vending

Features Products Operations Technology
AI-prevented food waste

How machine learning predicts demand and supply


August 4, 2020
By Naomi Szeben


Topics
Photo credit: ArtCookStudio/Adobe Stock

C anadian Vending and Of-fice Coffee Service spoke with Crisp’s founder, Are Traasdahl,

about the AI-driven program he created. Though Crisp was invented with the intention of reducing food waste, the pandemic generated a new use for Crisp: anticipating food shortages of finished goods, or scarcity of ingredients before a product was made. Crisp is a program that can inform an office coffee vendor of a delay or a com-plete absence of a certain product, immedi-ate decisions can be made to replace or wait for a delivery of key goods to be made.

The Harvard Business Review reported on the problems of supply chains in an ar-ticle published on March 27 this year, as North America shut down its businesses and urged consumers to stay home. The authors, Thomas Y. Choi , Dale Rogers and Bindiya Vakil reported, “a small minority of companies that invested in mapping their supply networks before the pandemic emerged better prepared. They have better visibility into the structure of their supply chains. Instead of scrambling at the last minute, they have a lot of information at their fingertips within minutes of a poten-tial disruption. They know exactly which suppliers, sites, parts, and products are at risk, which allows them to put themselves first in line to secure constrained inventory and capacity at alternate sites.”

Crisp offered that advantage, and re-ally showed its value as the early days of the pandemic introduced trucking delays, breakdown in supply chains and some-times, complete absences of much-needed products. Traasdahl spoke of his travel-inspired invention. While staying in New Zealand, he witnessed an apple orchard that had 70 per cent of its apples rotting in the field after the end of the season. Simi-larly, the opposite problem was happen-ing in India, he noted, and his mission to find out why began.

Advertisment

“We did research on this…we see that often more than a third of the food that’s being produced does not reach the consum-er at all. I’ve been aware of that for a time for a long time and we all try to do something about it. But one of the really big issues is all the food that’s lost be-fore it reaches the consum-ers,” stated Traasdahl.

Are Traasdahl, founder of CRISP.

According to Traasdahl, Crisp follows the supply chain all the way from through each steps of a supply chain. For many products such as coffee, many companies may be involved as well as the cost of imports and exports, roasting costs, packaging and finally, marketing.

“We capture data along the whole sup-ply chain, use that data to produce very ac-curate sales forecasts. We interviewed over 100 executives in the in the food industry, and it all came down to one thing: it’s in-credibly hard to have an accurate sales forecast; That’s the reason that sometimes we have too much; If you have too much you have to throw it out. Have too little, then you lose out on revenue and have up-set customers because you have too little.”

Traasdahl’s twenty years of technological experience with data mining came in handy. According to him, the use of the program is easy and can allow for a sales forecast that allows any baker to know how much they will sell in three months, or in a year. “That’s been very helpful in terms of the reduction of food waste but also in terms of creating more revenue.”

While office coffee service operators and pantry services experience peaks and valleys, the system uses a vendor’s own inventory history and past sales to isolate demand.

While holidays are one demand “driver,” events like pricing and promotions also influences the date. “What the software does is go back three to five years and it takes all that data. Crisp uses machine learning and artificial intelligence for lack of a better word, and then it separates out all of these different demand drivers so when you see the same combination, you know what to expect.”

“We typically say that we need at least two years of data. We’d like to have three years, and if we can get five years, great. But, if somebody has 10 years or 20 years, that makes our algorithms and our technology work even better.”

“We spent a lot of time to make this as simple as possible and easy to use. So, getting the data into this platform was very important. The OCS just creates a sales report that goes back the number of years they want and they send that into our system and we do everything on our end. We fit that into our system and then minutes later they can go in and get a forecast for it…typically some data clean-up needs to happen, first. There might be a situation where a company found 10 new offices to service, or there was some particular promotion that happened that created a big spike, or the OCS was missing one particular ingredient for a long time so they couldn’t provide that product.”

“So they typically kind of go through the history of the products and we identify for them. If these data lines seem like our players, and all of a sudden we have a big spike or a big drop. So then, the software allows them to put in. Then we take that into our calculations when we calculate the future.

“Most ERPs or at least the ERP systems that we work with, do not have great forecasting functions. So, we then do the forecasting based upon the data, we get from the ERP and then we can push it back so that the OCS or pantry service then has our accurate forecast. Based upon if they can do outsourcing or if they can do production planning; or if they can do labour planning, they can do all those things that the ERP system is built for; our forecasts are very accurate.”

The aforementioned Harvard Business Review article also asks important questions that a procurement office with the most advanced ERP might need to consider:

“When the procurement function has to resort to extraordinary measures to secure supplies on time (e.g., by expediting shipments or purchasing parts or materials at a premium), the higher costs are assigned to other parts of the organization (the logistics function in the case of expedited shipments, and the finance function in the case in the case of premium prices for raw materials and parts). Often, no one asks: Why was expediting or paying a premium necessary in the first place?

“People from procurement, logistics, and supply-chain financing need to come together to talk about what key gaps (tools, information, people, processes, etc.) need to be fixed to protect the company from disruptive events in the future and how to align the goals of procurement with the overall business objectives.

This promises to be of great help during the pandemic, when many vending operators reported a delay in deliveries or the supply chain.”

The global health crisis has showed everyone their weaknesses. Some companies have learned what their key products were, and what alternatives would and wouldn’t be accepted by their clients.

Other companies have set a crisis management protocol in place for procurement for office coffee supplies.

We’ve learned that companies who have survived did so with acute analysis and rapid communication.

Finding a system that can accurately measure inventory and predict needs so that a pantry service operator can communicate with suppliers should help improve the bottom line, and eliminate food waste.

Some may feel that machine learning has little place in a the OCS industry, but if the wild swings of availability during the pandemic have taught us anything, it is to look to our past to determine our future.