Farmers urged to question AI suppliers before signing up

Smart cow collars, camera-based locomotion monitoring, laser weeders and automated slurry scrapers are already working on many farms — but the president of the scientific-technical committee at the Italian Association of Artificial Intelligence Professionals (AIPIA) says farmers are signing up without asking the right questions.

Rafael Patron argues the core problem is simple: when it comes to AI, farmers rarely get a view from "someone with no stake in the sale". Vendors set the terms of the conversation. Farmers rarely do.

That matters because the answers shape daily herd decisions, data ownership and, potentially, relationships with milk buyers.

What should farmers ask before buying?

AIPIA — an independent non-profit based in Rome that brings together developers, data scientists, researchers and consultants, and keeps a public directory of AI professionals — wants farmers to interrogate exactly what a system can and cannot do.

Patron points to systems that use camera footage of housed cows to build an individual profile for each animal. The technology looks impressive. It has limits.

"It does not see the cow at grass in July," Patron said. "It does not examine a foot, and it cannot tell you why a cow's mobility has dropped."

His advice: ask the IT supplier which conditions the system is and is not designed to detect, and what happens to the score when cows are out for the summer.

False alarms deserve equal scrutiny. Patron said farmers should ask about them directly, because both failure modes carry a cost.

"A system that flags too many cows gets ignored within a month," he said. "One that flags too few gives a false sense of security, which is worse."

Who owns the data?

One issue that may not be exactly transparent, according to Patron, is who owns the information AI systems gather. Camera systems do not just collect animal-level data — over time they build a portrait of the whole business.

"Daily footage and scores for every cow add up, over a year, to a detailed record of how a farm is run: stocking, bedding, feeding routines, how quickly problems are picked up," he said.

That record can travel. If the monitoring system sits inside a buyer's scheme, the commercial implications are obvious. Patron's guidance is blunt.

"If the system is part of a buyer's scheme, find out whether the buyer sees the raw scores, a summary, or only what the farm chooses to share," he said.

He also urges farmers to ask where the footage is stored, how long it is kept, whether it is used to train the supplier's algorithm, and what they can take with them if they leave the scheme or change supplier.

None of this is necessarily a reason to avoid a system, Patron stressed. But he believes it is important to get the answers to these questions in writing — before installation, not after.

Who is in the loop?

AI connected to existing farm software can help, Patron said, but only if the scores land in front of a human with the time and authority to act on them. That needs planning.

"Before installation, agree with your vet which alerts will trigger a visual check and which will go straight onto the herd health plan, and decide who on the farm looks at the dashboard each day," he advised.

His view of who wins with these systems is pointed. Farmers who get the most from them "will probably be the ones who already know their cows well and use the scores to settle arguments with data rather than to replace their own eyes".

Patron's final recommendation is practical: visit a farm that has run a similar system for a full year — "ideally one that looks like yours". One full turn of the seasons, not a demo day, is the real test of any monitoring platform.

For farmers weighing a purchase, the checklist is short. What does the system detect, and what does it miss? Who sees the data? What leaves with you when you switch suppliers? Get it in writing.