Building a state-of-the-art data platform and supporting data infrastructure for a major tomato producer
Agrocare
AI consultant in Den Bosch
For growers, processors and suppliers in north-eastern North Brabant: we build data platforms for BI and data science, as we did for tomato producer Agro Care.
A data platform to make all of Agro Care’s data centrally available for BI and data science
Agro Care, one of Europe’s leading tomato producers, was ready for the next step: a scalable and stable data environment. Our brief: a data platform with a data warehouse, using Databricks.
When a supermarket raises a complaint, how do you quickly identify the greenhouse, production line and batch?
A retailer reports an unusual colour or a shorter shelf life. The quality manager looks up the batch number in the ERP, the sorting data on the machine and the lab result in a spreadsheet. We identify which integration can eliminate that search first.
Cultivation, production line and lab data only make sense when brought together
One platform for all data
Climate, harvest, sorting, orders and lab results in one place. The brief at Agro Care: centralise all data, ready for BI and data science.
Harvest and sales forecasts
A model based on cultivation history, weather and orders proposes volumes, so planning and sales teams know sooner what is coming in.
Inspection using camera images
A computer vision model assesses colour, shape and damage on the sorting or packing line and links rejected items to their batch.
Management after go-live
Pipelines, monitoring and a designated owner keep figures and models accurate when a new variety, product or customer specification is introduced.
We trace a single batch from greenhouse or raw material to the customer
-
Trace one batch
Together with the cultivation, quality and planning teams, we trace one batch through the climate control computer, sorting machine, ERP, lab and complaints log.
-
Test one question
Rejections, shelf life or the volume forecast: using batches from previous seasons, we test what the model predicts correctly and where it gets things wrong.
-
Connect each line or site
What works runs on Databricks or Fabric. The next line or site connects to the same batch data.
Is this a good fit for a grower, processor or supplier?
A good fit if
You grow, process or supply products in the food chain. Your climate control computer, sorting machine, ERP system or lab already records data, but quality or planning still depends on searching through separate systems.
Not a good fit if
Batches, rejections and complaints are not yet recorded, or are only recorded on paper. A model would have too little data to learn from. The first step is to improve record-keeping and agree who records what.
What determines the scope
The number of sources and locations, how many seasons of historical data you have, whether a platform already exists, and who will manage it: your team or ours.
Too few seasons of historical data?
A harvest forecast learns from previous crops. If only one or two seasons have been recorded, start with the platform and dashboards. The model can follow as you build up more historical data.
What comes first: the data platform, a forecast or image-based quality control?
Start with the data platform
Choose this if cultivation, production line and order data are spread across separate systems and spreadsheets. Our advice: start here if every question first requires a search for the data. See our data and AI platform.
Start with a forecast
Choose this if your data is already brought together and your planning or sales team wants to anticipate harvests or demand. A data science model can then provide a forecast.
Start with image-based quality control
Choose this if people inspect products visually on the sorting or packaging line and there is room for a camera. Read what computer vision requires to do this.
The data platform and the models running on it are built by the same team
Since 2016, Data Science Lab has helped organisations work with data and AI. As a Databricks Partner and Microsoft Partner, we design the platform and build the models that run on it. Our colleagues test new ideas in our internal lab.
We’ll choose one product and one question together in our first conversation
Stef Stoelwinder
Business DevelopmentStef Stoelwinder is your first point of contact. In our first conversation, we’ll discuss the sources in the greenhouse, on the production line or in the lab, and choose the first question to address. We can meet at your premises, at our office in Amsterdam or online.
E-mail info@datasciencelab.nl
Phone +31 (0)20 3636 724
Which batch, production line or complaint are we still investigating manually?
Describe your question about cultivation, processing or sales. We’ll reply by email or arrange an initial conversation, at your premises, ours or online.
Message sent
Thanks for your message. We’ll be in touch as soon as possible.
Frequently asked questions
For growers, data comes from the climate control computer, including temperature, light, irrigation and energy use; from crop records; and from the sorting machine, which records weight and grade for each batch. For processors, it comes from the ERP system, production line controls, the laboratory and HACCP records, which document food safety checks.
These sources often run on systems from different suppliers, sometimes in the machine manufacturer’s cloud. The first step is to find out what you can export or retrieve through an API, and how often.
The EU General Food Law Regulation (178/2002) requires you to identify at least your immediate supplier and immediate customer: one step back, one step forward. If you also want to know which greenhouse, production line and raw material a product came from when a complaint arises, you need to link batch numbers internally.
On a data platform, the batch number links crop records, the sorting machine, ERP, lab results and complaints. The main challenge is usually not the technology, but agreeing who records which number and when. That is part of data governance.
A computer vision model learns from labelled examples: accepted or rejected (classification), or the location of damage (detection and segmentation). Whether it works depends on the difficult cases: wet products, glare, a new variety or different packaging. These need to be in the training and test sets.
Decide in advance which error is worse: missing a defect or rejecting a good product. After deployment, someone regularly checks a sample so you can spot when the model starts to drift. We keep a model like this running through management and continuous improvement.
Yes, if enough comparable seasons have been recorded. A harvest volume model uses planting date, variety, light, temperature and the harvest so far. For sales forecasts, it also uses orders, retailer promotions and the weather. To see how retailers approach demand and stock, read about data and AI for retail.
Compare every forecast with a simple benchmark, such as the same period last year. If the model cannot beat it, it is not yet an improvement. There is little historical data for a new variety, so have the grower review the forecast as a check.
Our site does not feature a case study of a company from Den Bosch or North-East Brabant. One of our agrifood case studies is Agro Care, one of Europe’s leading tomato producers, headquartered in Wateringen.
We do not have an office in Brabant either: our only office is on Westerdoksdijk in Amsterdam. We can meet at your premises, at ours or by video call.
The Jheronimus Academy of Data Science (JADS) is a partnership between the province of North Brabant, the municipality of ’s-Hertogenbosch, Tilburg University and TU/e, with its Mariënburg campus in Den Bosch. Under its AgriFood & Nature theme, JADS brings together research, education and businesses.
Data*Grow was launched there on 28 January 2026 by Growers United, LetsGrow.com, Open Universiteit and JADS. It focuses on data in high-tech fruit and vegetable growing. That is where research and training take place. We build the application for your own business and keep it running in production.
AgriFood Capital, North-East Brabant’s regional development agency, is based at Agri&Food Plaza in Den Bosch. Together with Brainport Development, it runs Next Tech Food Factories, where food processors, high-tech suppliers and machine manufacturers develop and test digital technology such as smart sensors, AI and robots.
MKB Team Noordoost-Brabant works with entrepreneurs to draw up an action plan covering areas such as data, AI and cybersecurity. Programmes like these help you explore possibilities, run tests and find partners. Our role is to build the platform or model that then runs in your own process.
When the problem lies in how data is recorded. If you enter batch details by hand or too late, a model will only give you wrong answers faster. Start by improving data entry. If you mainly need to know what happened yesterday, a Power BI dashboard using existing data is enough.
If a scale or colour sorter already measures what you need to know, you do not need a camera and model. And if you only want to explore whether something is technically possible, a regional programme’s field lab may be a better fit than a consultancy project.