London Produce Show: How Fresh4cast is helping the fresh produce industry get smarter
From the pages of Produce Business UK
Providing forecasting solutions for leading fruit producers such as Berry Gardens and Angus Soft Fruits, Fresh4Cast has become one of the leading fresh produce supply chain management tools. It will be exhibiting at this year’s London Produce Show, which runs from June 5-7 at Grosvenor House, and says one of its primary aims for the show is to try to expand its customer base, particularly when it comes to upping its presence around salads and beans.
We caught up with Fresh4Cast chief executive Mihai Ciobanu to ask how the business plans to evolve and understand whether Brexit is changing its business model.
What inspired Fresh4Cast as a business?
During my Sloan Fellowship in 2013, one of my colleagues opened my eyes about the challenges of the fruit and veg sector. His family had been in this business for many years, and despite having many experienced people in the team, they still had a hard time forecasting their harvest and the market demand. The more I looked into this, the more I understood how generalized the problem was.
This lack of predictability creates volatility and inefficiencies, including significant waste, and puts enormous pressure on sustainability. But these types of problems had been solved in other sectors already. Although the challenges in fruit and veg are indeed very specific, even compared to other sectors of agriculture, it was possible to build an economic solution. I was looking for purpose, for a way to make a difference in the world. So I tried to surround myself with the right people (of course starting with the original colleague who became my co-founder), and we got to work.
How exactly do you help UK businesses, specifically soft fruit growers?
The foundation of good forecasting is good data, so the first thing we do is help our customers integrate all the relevant data flows into a single source of truth. This starts with internal data, such as production plans, harvest records, sensor data, sales data, market data and so on. For convenience, we add external information, such as weather, import-exports and exchange rates. All of this information is streamlined, matched up and presented back to the users in specialized visualization tools.
Automated forecasting tools form the core of the service. These are programs that use machine learning to model various business processes and make predictions at a level of speed, detail and accuracy that would be very expensive to match otherwise. Our customers’ forecasting managers, production managers and sales managers effectively get a small army of software robots to help them in their work. Compared to human forecasting alone, these tools can reduce errors by more than a third. Finally, all these data streams are integrated into the customers’ IT systems. The forecasting models can be used to run scenarios, which can then be pushed into the MRP (material requirements planning), an Excel file or some other tools.
But can there ever be too much mechanization? Is it a worry that perhaps too much faith will be placed in machines over human beings?
If we take a wider view of our society today, you can say we are all cyborgs to some extent. We form a near-symbiotic link with our phones and computers for basic tasks like communication, finding our way around and so on. And as long as we use the time gained for more valuable tasks, there is significant value in this. A similar trend can be seen in agriculture. Each company now has a myriad of data sources, and with the right tools, they can turn those into assets. As with humans, there is a balance between trust and checkpoints. The same way checks and processes are put in place for human work, it will be increasingly important to have checks in place for machines.
Even if some of these will be inevitably performed by yet more computers, it means that for critical tasks, companies can avoid having a single point of failure. An example of human checks are company accounts, which are typically signed off by at least two people before being filed. An example of machine checks is redundancy: for some of our customers, we provide two independent production forecasts running side by side, with completely different algorithms and input parameters. And these are both compared against the growers’ own forecasts. The key to reducing risk with advanced technology is to adopt it gradually.
What would you say are the biggest challenges in the fresh produce business right now? And how are you helping fresh produce brands overcome them?
Complexity is increasing due to consumer trends, advances in production, storage and transportation, marketing, international trade and not least, weather patterns. This increases volatility, which in turn puts pressure on the time horizons of the decision-making process. But short-term thinking is not good enough in a context of large shifts in global trends, regulatory changes, water scarcity and other constraints. We try to help with forecasting models and scenario planning tools, which facilitate analysis and conversation.
Supply chain management can be a force for good, particularly in terms of reducing waste and seeing what's going wrong in your supply chain. Can it help fruit producers be more sustainable in terms of their environmental impact?
Multiple studies show that one-third of all fruit and veg produced worldwide every year is wasted. This is consistent across developed and developing regions; the only difference is that developed countries have efficient supply chains which push the problem downstream. When a rural market in India faces oversupply, the produce is written off close to the source. When the UK faces oversupply, the produce is pushed to consumers with short-term price promotions. Often, the same consumers buy more than they need and the surplus ends up in the bin at home.
So how can this be improved? Well, when marketing companies have a few more weeks more notice about an oversupply, they can plan more efficient promotions with their retail partners, using communication and improved in-store visibility to increase penetration rather than basket size. Demand forecasting models help them further by allowing them to calibrate the investment to match the product availability. Yield forecasting combined with demand forecasting facilitates harvesting decisions. On rare occasions, foregoing part of a crop, painful as it is, can be much more profitable than harvesting, packing, transporting those products and then crashing the market prices. And in more regular circumstances, actions can be taken to speed up or slow down certain crops in order to fine-tune the balance between availability and demand. All of these decisions can reduce waste, improving both the environmental and financial sustainability of this crucial sector.
What are some of the big technological innovations Fresh4cast is currently working on?
In the near term, we continue to focus on our forecasting solutions, using the latest deep-learning technology. Our typical demand model, for example, uses one polynomial regression and several competing neural networks. These bespoke learning applications are the core of our services. In the medium term, significant value can be generated by integrating different tools. On one hand, building the ecosystem with other IT solutions; on the other hand, integrating more user input in our existing customer apps. We are exploring a few ideas with customers now, and we are delighted with the response.
What are some of the biggest mistakes fresh produce businesses make when it comes to supply chain management? And how can they be overcome?
I would not presume to draw a general conclusion about mistakes, but many companies say their biggest challenge is forecasting. When a customer starts to work with us they often tell us “our forecast is usually ok, but sometimes it goes terribly wrong.” The first revelation comes when they see their accuracy measured correctly. Every process improvement begins by measuring the right key performance indicators.
So that’s always our approach at the beginning: we unify all the relevant data and present it back to the customer, with the right measurements in place. We also estimate the accuracy of the initial forecasting models, so that the customer can make an informed decision. Finally, we recommend a roadmap where the customer app is deployed in stages, evaluating the success at least once a year. This is fundamentally different from traditional software decisions, which used to start with a multi-year process of drawing up “the ultimate” specification, followed by an even longer deploy process. We believe that a lean, gradual build-up is much better. Each step is driven by user adoption and engagement and therefore generates much higher returns on investment.
Where do you see the business growing to in the future?
Today we help forecast 67% of the UK strawberry production, and we have a growing coverage in other categories and geographies. Our demand forecast also has a growing coverage in the UK, and we are preparing the first projects abroad. In 10 years, I’d like us to forecast 20% of the volumes of all fruit and veg produced or sold in Europe. Importantly, I want to keep our services affordable, so that small and medium-sized companies can continue to get the benefits as they do today, with no capital expenditure and low annual fees.