By the YesPress Desk / Filed from Potsdam, Brandenburg / Crop Growth & Yield Prediction
Here is a fact about food that nobody puts on a menu: a lot of it is wasted simply because no one knew how much was coming. A grower plants a greenhouse full of tomatoes, tends them for months, and then - roughly - guesses when they will ripen and how many there will be. Buyers want commitments earlier than that guess can honestly support. Energy bills arrive whether the plants cooperate or not. Labor has to be scheduled against a harvest nobody can quite see. The whole thing runs on educated intuition, which is a polite way of saying it runs on hope.
HarvestAi, a company of about twelve people headquartered in Potsdam, Germany, is trying to replace the hope with arithmetic. Founded in 2020 and led by CEO Dr. Georg Caspary, the company builds software that forecasts crop growth and yield inside high-tech greenhouses. The product is not a robot and it is not a drone. It is a web-based platform that ingests harvesting records, climate data, autonomous camera feeds and external weather inputs, and then tells a grower - with a reported accuracy above 90% - when the fruit will be ready and how much of it there will be.
The elegant part is what that number does downstream. If you know your yield three weeks out, you can negotiate the sale three weeks out. You can staff the harvest correctly instead of over- or under-hiring. You can decide whether it is worth running expensive climate control during a high-energy period, or whether the plants will get there anyway. One good forecast, it turns out, quietly solves several unrelated problems at once. That is the sort of leverage that makes a twelve-person company interesting.
HarvestAi's own framing is less about robots replacing growers and more about giving growers a crystal ball they can actually act on. The team deliberately mixes disciplines - AI engineers alongside plant biologists and physicists - on the theory that a tomato does not care about your machine-learning architecture, and you cannot forecast a plant you do not understand. The model learns the biology first, then the software gets to be clever.
The proof, as always in agriculture, is in the dirt. In a collaboration announced in early 2024, HarvestAi partnered with Gemueseproduktion Zorbau to run its forecasting technology on a commercial tomato operation of more than eight hectares in Lutzen, Germany - an area larger than eleven football pitches. There the system does the unglamorous work: registering ripe and unripe fruit in real time via computer vision, folding in climate and weather data, and letting growers simulate different climate scenarios to see how each one would bend the harvest curve before they commit to it. Machine learning alone gets the forecast past 90%; the cameras push it further.
That distinction - full computer vision versus machine-learning-only - is also how HarvestAi meets customers where they are. Not every operator wants to install autonomous cameras across their glasshouse on day one. So the company sells the forecasting brain with or without the eyes, a pricing choice that reads less like a product-line decision and more like a company that has spent real time listening to growers who count their capital carefully.
