The fundamental challenge in food manufacturing is simple: no two raw materials are exactly alike. Just like potatoes, individual items such as peas, vegetables, and sugar can differ significantly: not just in size or color, but also in skin thickness, moisture content, or starch levels. This inherent variety makes it increasingly difficult for food manufacturers to maintain stability in their output quality, especially while under pressure from numerous stakeholders. Understanding and managing this input variability is what will distinguish which food processors become champions in their field.
The specifications of resources are crucial information for a food processor to set up their manufacturing process. Think of the size of tubers, color variations and defect occurrences on a carrot, or the fat content of chocolate. These resources inherently display a large variety in these specifications since food manufacturers work with organic products. It makes it impossible to handle every input batch the same way. This is very different from what, for example, automotive manufacturers are used to. In an ideal world, process parameters should be adjusted continuously to compensate for the changes on the input side. In practice, this seems infeasible, resulting in non-conforming products and a strong loss in process yield. As food systems face new pressures, this variability is only intensifying.
This has long been a challenge, but the speed and magnitude of change is something we've never seen before. Seasonal variations now introduce significant swings in resource compositions. For example, studies of milk show that fat, protein, and other specifications exhibit serious seasonal cycles. Early harvest crops often carry higher moisture levels and lower dry matter levels than late-season ones, meaning fixed processing parameters give very different outputs over the year.
On top of seasonal differences, food producers nowadays source their crops from a larger portfolio of suppliers to mitigate supply risks and keep costs stable. However, each region has unique soil, rainfall, and storage conditions which again increases high variation of raw inputs.
Finally, climate change compounds all this. In recent decades, droughts have accounted for over 65% of agricultural losses which translates to an estimated 3.3 trillion euros of loss in crop and livestock production. Global modelling even indicates that it has suppressed agricultural productivity by roughly 21% since 1961, washing away nearly a decade of gains in productivity. As weather conditions become even more volatile and growing conditions even more difficult, those variations will for sure also transfer to the crops that you want to process.
The cost of input variability might be hiding in plain sight, but it's far from negligible. It nibbles away at profit margins every day. In fact, Eurostat estimated that up to 20% of its food losses occur at a food manufacturing plant. Falsely configured optical sorters can't handle color variations, peelers that are set up too aggressively to compensate for shape irregularity, or ovens that need adjustments to handle different batches of product.
Waste occurs in all steps of processing. Now, consider a large potato processing plant producing tens of thousands of tons per year. Even a 1% loss in yield translates into hundreds of tons of lost products, translating to hundreds of thousands of euros per year. But it's more than a financial impact; variability also impacts the brand, which can result in customer complaints or even worse, recalls.
Moreover, each corrective action costs time, interrupts throughput, and introduces inefficiency. A plant operating even 2–3% below theoretical yield may lose more in opportunity cost and hidden waste than many realize. In short, input variability is a silent killer for the bottom line.
For decades, food production, like in other FMCG processes, was relying on stability. Fixed recipes, static setpoints, and operators who know how the line should sound, smell, or look when it’s running right. In bakery processes, to give an example, true bakers used to control the line knowing the ins and outs of the dough and its rising process.
Today's plants face more pressing issues such as the war for talent, but also higher throughput volumes, making it impossible to stick to this traditional way of working. Recipe-based automation and periodic lab feedback work well when raw materials behave consistently, but in modern plants, that's rarely the case. Nonlinear relationships between variables like moisture, starch, or viscosity mean that even well-tuned control loops struggle to maintain optimal performance when inputs shift.
Most PLCs and SCADA systems operate on fixed setpoints. They can't detect when the underlying process model no longer matches reality. Operators compensate manually, adjusting fryer temperature, mixing time, or feed rate by experience, but by the time the deviation is visible, the loss has already occurred. In effect, traditional automation delivers static optimization in a dynamic environment, leaving factories perpetually one step behind their raw materials.
If traditional control is static, the next generation of manufacturing must be adaptive. Instead of assuming that inputs are constant, adaptive processing continuously senses, learns, and adjusts. Just like an experienced operator would, but in real time and on an industrial scale.
These systems use data from vision, sensors, and inline measurements to understand what's truly happening in the product, not just the machine. When starch levels rise or moisture drops, parameters like fryer temperature, mixing speed, or drying time are automatically tuned to compensate. It's no longer about rigid recipes, but about dynamic control that responds to the product's behavior as it happens. Over time, AI-driven models can even predict variation before it affects quality, turning process control into process intelligence. The result: less waste, stable quality, and a higher process yield.
Let's dive a bit deeper into a practical example: the peeling process of potatoes. Whether it's French fries, crisps, or ready-made potatoes, they all use a similar process to remove the golden crop's skin. Either by using abrasive peelers or steam peelers. In the former, detection of skin remains after peeling can give an an indication of how peel time and peel speed should evolve to compensate for the changes caused by, for example, a switch in supplier. That allows us to consistently maintain the desired peel effectiveness, whereas manual adjustments by operators would require thorough process knowledge and objective and timely quality insights. Also, for steam peelers, similar adjustments could be made using brush speeds, steam pressure, etc.
To conclude, it's safe to say that input variability has become the operating reality of modern food production. Weather shifts, seasonal differences, and a broader supply chain add complexity to an already hard-to-control process. The plants that thrive will not be those that chase stability, but those built to adapt. They can capture the intuition of skilled operators and scale it across every shift, every product, and every season. In a world where variability is inevitable, adaptability becomes the new measure of control. The question for every food producer is simple: are your processes designed for yesterday's conditions or ready to perform under tomorrow's uncertainty?
