Cutting water and product losses starts with real-time data

Different potatoes, identical fries 

If you ask Yarne De Munck, co-founder and CEO of Polysense, what causes most food waste, he points to the variability of the raw materials: ”Size, moisture, sugar content: you never have the same potato twice. And yet, you can eat fries at a McDonald’s anywhere in the world, and they’re always exactly the same. To be able to deliver that identical size and taste, lots of potatoes that don’t fit the bill, get lost.” 

The reason: the processes that handle the potatoes run on settings that are tuned once and are rarely revisited by whoever’s on the floor that day. When the settings don’t adjust to match, some potatoes end up too big, too small, or off in some other way. Customers, retailers, and foodservice buyers have strict standards, and anything that doesn’t meet these gets rejected.  

The same goes for fruits, veggies, biscuits or any other food product. The quantity lost in the value chain is staggering, and every food manufacturer feels it directly. It is product they’ve paid to grow, and process, and then pay again to dispose of.

The hidden waste in water

Steven De Schrijver, co-founder and CTO of Liquisens, points to another raw material where losses show up during food production: water. “Water is less obviously linked to waste, but it is a precious resource for the food industry,” he says. “Beyond its role as an ingredient, water runs through nearly every step of food production: rinsing, steaming, peeling, and cooling. That are a lot of touchpoints where water can turn into waste or even cause waste elsewhere in the process.”  

“Take CIP, the cleaning-in-place cycle that runs between production stages. On paper, it follows a set process: use the exact amount of water, choose the right temperature, add the required chemicals, and respect the timing. But following these steps is no guarantee that the cleaning will actually work. The only way to know for certain is a lab result that takes a day or two to come back. That means that issues are only discovered when it’s too late to prevent them, resulting in rejected or even recalled products.”

Make water quality predictable 

Liquisens monitors water quality and builds predictive models to check, in real time, whether a cleaning cycle actually does its job. Steven explains: “Liquisens Predict draws on process data the plant already has, like temperature, the product that ran before, how long the cycle lasted, pH, conductivity, organic carbon readings, etc. to understand when a CIP cycle is successful for a particular line. Based on those insights, it flags a deviation before it becomes a problem.”  

That’s how Liquisens prevents losses: “Operators are used to building in safety margins in CIP processe. They make sure to use enough water and chemicals and to rinse long enough, just in case. Give them real-time data instead, and that margin can shrink without adding risk. With our tools, they get a signal while there’s still time to act: cleanse again if it’s really needed, hold a batch on purpose instead of out of caution, or move on with confidence. That avoids lost batches, or worse, costly shutdowns, and it means less water and fewer chemicals used overall.” 

Eyes on every potato

Polysense works on the same principle: real-time visibility instead of a delayed result, just aimed at a different part of the line. Where Liquisens reads and predicts what's happening through process data, Polysense relies on visual detection, watching every single item that crosses the line. 

The solution works in two steps. First, the in-line quality control is fully automated: every product gets measured objectively (size, shape, defects, …) and continuously as it moves across the line, instead of an occasional manual check here and there. “Vision systems do the inspection,” Yarne says, “and we process results locally, so that the data is available in real time. To train the system to handle dozens of product types without months of manual labelling, we build much of our training data synthetically. We simulate the production line in a 3D environment to generate the range of variation a real camera would eventually need to recognize.” 

That data then feeds straight into automatic process control: machines and parameters adjust as conditions shift. The result is stable output quality that meets what customers and retailers expect. Waste goes down, and production teams face less pressure to catch every deviation by hand.

Not just theory

Polysense’s platform runs across a growing number of food manufacturers. The savings are substantial. “One of our customers, a producer of frozen fries, used to check product dimensions through manual sampling,” Yarne says. “That left a lot of room for inconsistency from one operator to the next. Since switching to continuous, automated inspection, they can catch deviations across their entire range of products quickly and accurately.”   

Steven sees a similar pattern at Liquisens’ customers. Just like Yarne, he’s cautious about sharing ROI and figures, but the potential is real: industry research suggests that the optimization Liquisens works on can realistically cut chemical use by around 24%, while energy costs too drop (-17%).

More than the sum of two