Why start now
The Spanish fruit and vegetable industry has spent years doing things reasonably well with spreadsheets, vertical ERPs and a great deal of field experience. What has changed over the past 18 months is that regulatory pressure and cost pressure have arrived at the same time. Royal Decree 1051/2022 mandates a digital field notebook. The CSRD mandates environmental footprint reporting. Fertiliser prices remain high. And end customers no longer buy without traceability.
A mid-sized cooperative coordinating between 60 and 400 growers cannot take all of this on with people alone. Nor can it start a monolithic 18-month AI project — because the results arrive too late.
What does work: identify five specific processes that can be automated in a quarter, measure the impact, and scale from there.
01 · Optimising intake
Intake is the classic bottleneck. Members bring produce in concentrated windows, the plant saturates, queues build up and temperature losses start appearing before anything reaches the grader.
What AI automates here is the hour-by-hour forecast: when each batch will arrive, at what expected quality, and how much plant capacity is available. The final decision stays with the season manager. The model feeds on field notebook data, weather reports and the record of previous seasons.
Typical result in 12 weeks: -22% waiting times and +18% capacity utilisation. It requires no new hardware. It does require the field notebook to be digitalised — that remains point zero.
02 · Season planning
Until now, season planning has run on three inputs: history, intuition and the February conversation with members. AI replaces none of that. What it does is add a fourth dimension: a forecast by member, batch and expected quality, with an honest confidence interval.
The sales team can then sell 30 days earlier with firmer numbers. And the agronomy team can prioritise visits to the plots most at risk of drifting off target. The difference from a spreadsheet: the model learns season by season instead of staying frozen in 2019.
03 · Digital field notebook
Royal Decree 1051/2022 is already mandatory. What many cooperative managers have not gauged is that the digital notebook is not an end in itself — it is the raw material for everything else. If your members are still writing down doses by hand, you will not be able to use AI seriously.
The good news: with an app designed for rural mobile use (offline-first, in Spanish and Valencian/Catalan, syncing once back on WiFi), real adoption at 8 weeks reaches 90% if the cooperative pushes for it. What does not work: imposing the notebook as an administrative obligation without giving the member anything back. The app has to help them — dose recommendations, weather alerts, deadline reminders — and not merely collect their data.
04 · Yard quality control with computer vision
This is where visual AI genuinely comes in. A fixed camera in the yard or on the grading belt, a model trained on the defects of your own vertical (splits, bruising, pest damage, over-ripeness), and the subjectivity of manual inspection drops sharply.
A warning: generalist models do not work. You have to retrain on images of your own produce, ideally labelled by your own quality technician. It is 4-6 weeks of work the first time, but after that the model scales to other seasons with very little effort.
Comparable cases we see: +15% average certified quality on the line, which translates into a better price at destination and fewer complaints from the end customer.
05 · CSRD reporting by default
The CSRD directive affects large cooperatives directly and, in cascade, their suppliers. Reporting carbon, water and energy footprint per kilo produced is now an expectation rather than an option.
Automation here is relatively trivial if field and plant data are integrated: the system cross-references production, energy consumption and agronomic applications, and returns a report that stands up to scrutiny. What is not trivial is having the data in the first place. That is why this is block 5 and not block 1.
Conclusion
If your cooperative meets both conditions — a team willing to digitalise the notebook, plus a commercial decision to measure better — the five blocks fit into a quarter. Not into a single month, but into three with reasonable parallel working.
The most common mistake we see: wanting to start with computer vision (the most demo-friendly piece) without having digitalised the notebook first. It is like fitting Tesla Autopilot to a car with no GPS.
30 minutes with our team. Zero slides.
We tell you exactly where AI moves the needle in your operation — and where it does not.
Book a demo→