Source.ag Expands AI Harvest Forecasting to Pepper Cultivation

Source.ag’s expansion of its AI Harvest Forecast to pepper cultivation addresses inefficiencies in greenhouse horticulture by reducing yield prediction errors by more than 40%. This automated integration of climate and plant data allows growers and sales teams to stabilize supply chains, mitigating the 13% global post-harvest food loss cited by FAO. For high-tech agricultural operations, these advances facilitate improved labor management and retail commitment accuracy in an environment of rising operational costs. 

_____

Amsterdam-based data and AI platform Source.ag announced the commercial launch of its Harvest Forecast solution for pepper cultivation. This expansion follows the technology’s application in the tomato sector and represents the first time the company’s automated forecasting models have been adapted for pepper varieties.

The AI-powered system replaces manual Excel-based yield calculators previously used within the Source Platform. By integrating greenhouse climate data, cultivation strategies, and specific plant information, the model generates automated weekly predictions with a forecast horizon of up to eight weeks. Initial support covers six block pepper varieties, including Alzamora, Gina, Cadalora, Yedda, Frazier, and Silverstone, with further varieties expected to be added in the near future.

The company notes that pilot results from growers at Harvest House and Jansen Paprika indicate that the AI model reduced average forecast errors by more than 40% at the three-week horizon. The data also showed a 28% decrease in outlier weeks, where manual forecasts typically deviated by significant margins. According to cultivation managers at these facilities, the automated process reduces the administrative burden of manual data entry and provides more reliable data for labor scheduling and retail supply commitments. "In pepper growing, you always want to know what is coming. If you can predict that accurately, you can manage more effectively and efficiently, and align better with sales and planning. With labor costs continuing to rise, more efficient planning is no longer a luxury, it is a necessity,” said Gerard Kremer, Cultivation Manager, Jansen Paprika.

The transition from tomato to pepper forecasting required distinct adjustments by the company’s data and plant science teams due to the different biological characteristics of the crops. The system is designed to update automatically as new information becomes available, requiring head growers to input only round-specific and fruit coloration data on a weekly basis. This streamlined reporting aims to provide sales and logistics teams with more consistent data to manage supply chains amid rising labor costs and fluctuating retail demand.

Development of Next-Generation AI Models

The development of a pepper-focused model follows a breakthrough in Source.ag’s next-generation AI forecasting for tomatoes announced on Mar. 19, 2026. The company said this iteration introduced fundamental changes to how the model is trained and updated, utilizing real-world cultivation data to account for climate variability that individual growers may not encounter in isolation. To achieve this, Source combined horticultural science with AI that adapts to the specific setup and ecological complexities of each greenhouse. The model is designed to improve performance as more data is processed across the platform.

According to Rien Kamman, CEO and Co-Founder, Source.ag, the model's performance compounds as more growers use the platform, creating AI that improves with scale. A primary operational objective of this update was to reduce manual registrations. "The most tangible change for growers is how much less the model asks of them," said Sebastiaan Vermeulen, Data Scientist, Source.ag. He noted that automating data processes removes friction from the grower’s week, while changes in data utilization lead to improvements in the prediction accuracy. Vermeulen added that these advancements facilitate hitting harvest commitments, realise better prices, and reduce waste.

A primary operational objective of this update was to reduce manual registrations. Several data processes were automated to remove the recurring administrative burden while simultaneously increasing prediction accuracy. Field testing results for this tomato-specific model showed a 33% increase in mean forecast accuracy at the three-week horizon compared to previous versions. Additionally, severe outliers, defined as cultivations with significant forecast misses, were reduced by 50%, while the percentage of cultivations missing targets by more than 20% declined by 25%. According to the company, these improvements are intended to help growers meet harvest commitments, realize better pricing, and reduce waste.

Food Loss: A Pain Point for the Global Food System

Reducing food loss is critical for improving food security, promoting resource efficiency, and protecting the environment. Currently, over 13% of food is lost globally in the supply chain after harvest and before the retail stage, according to the UN Food and Agriculture Organization (FAO). The organization notes that an additional 19% is wasted at the retail, food service, and household levels. Combined, food loss and waste account for an estimated 8 to 10% of global greenhouse gas emissions.

The OECD-FAO Agricultural Outlook Report projects that halving food loss and waste could reduce global agricultural greenhouse gas emissions by 4% and the number of undernourished people by 153 million by the year 2030. To achieve these targets, increased climate investments and the redesign of storage systems to reduce upstream supply chain losses are required.