Advancements in generative AI have been heralded as a new age for businesses across every sector. In dairy, there’s been a rush of software vendors promising transformational results, but there’s a wide gap between the pitches and the outcomes delivered.
Today, "AI-powered" can describe anything from an established machine-learning model solving a targeted operational problem to an agentic system with the authority to take actions and change how a business operates. Fundamentally different capabilities, with different benefits and risks.
How can dairy executives move quickly enough to capture the opportunities without getting caught up in the hype?
The answer begins with the same questions you'd ask of any significant investment. What problem are we solving? Is this the right solution? And crucially, what are the risks?
Start with the problems, not the technology
We haven't spoken to many dairy leaders who are kept awake at night wondering how to deploy more AI. They're thinking about things like milk supply, quality, input costs, farmer retention, efficiency and risk. These are business problems, not technology problems.
Established forms of AI are already solving specific problems and delivering operational benefits. Generative AI needs to prove its value: how does it solve a real problem better, faster, more reliably or more economically than anything else?
MADCAP's origins date back to 1974, when it began working with the New Zealand Dairy Board to connect dairy analytical equipment with computer systems. We’ve been evolving and automating complex processes ever since, embedding decades of data, business logic and operating rules in the systems processors already depend on.
The opportunity with generative AI isn’t to replace the foundations that are working well; it’s to make them more accessible and useful. In tech circles, there’s a saying, ‘there’s no AI without APIs’; in other words, a well-architected core system with safe bridges to share data with other applications now matters more than ever.
Where generative AI can make a real difference
Consider exception management. Systems are already capable of reviewing results, identifying deviations from business rules and alerting an operator. Using an AI model to replicate that proven logic adds complexity without necessarily adding value. The more interesting opportunity is what happens next.
-
Imagine an operator asking, in plain language, why milk supply in a particular region has fallen compared with the previous fortnight. Rather than waiting for someone to assemble reports across several operational systems, they instead get a simple explanation drawing together relevant collection volumes, supplier movements, quality results and other available operational information.
-
Or a commercial manager asking which suppliers and collection patterns are contributing most to increased transport costs, with the ability to explore the underlying information and test alternative scenarios.
That’s just two examples, and they highlight an important distinction in how generative AI should work. Generative AI should never establish the operational facts on which its answers depend. These need to come from your trusted systems of record capturing milk movements, quality results, supplier agreements, payments and business rules. The AI layer provides a more flexible way to access, combine and understand them while always clearly signalling the difference between fact and possible explanation.
The opportunity is not to create an artificial source of truth. It is to make the information the business already trusts more accessible
What happens when AI moves from answering to acting?
Answering a question, recommending an action and executing that action are not the same thing.
An AI assistant that explains why a producer payment has changed presents one kind of risk. An agent authorised to change payment rules or release funds presents another.
That doesn't mean AI should never take action. There are opportunities in dairy to automate more of the routine decisions and exceptions within carefully defined boundaries.
The question is how much authority the AI should have, and where that authority is enforced.
Imagine an AI capability that investigates an exception, retrieves relevant facts, checks the applicable business rules and recommends a resolution. Routine cases meeting predefined conditions could be automatically executed under clear controls, with more complex cases escalated for human review.
The AI can help interpret intent and prepare an action. It should not be able to bypass the controls that make the underlying business system reliable.
The risks go beyond getting the answer wrong
There is a familiar principle here: poor-quality information produces poor-quality decisions. Generative AI doesn't eliminate that problem; it can actually make it harder to recognise by presenting an incorrect interpretation in convincing language.
Processors also need to know where sensitive information is processed, what the model can access and retain, whether their data is used for model training, where the infrastructure resides and who is accountable for resilience and cybersecurity.
These are not reasons to avoid AI. They are reasons to maintain the same standards of due diligence that should apply to any technology supporting business-critical operations.
What is the lowest risk option?
For us, one particularly promising approach is Private AI. Generative AI operating within defined security boundaries that govern information access, processing, retention and use. Customer information remains protected and inaccessible for model training without explicit authorisation.
That is the architectural direction we believe deserves serious investment and evaluation, with its actual benefits and costs measured rather than assumed.
The emerging opportunities with generative AI could make a meaningful difference to how dairy businesses operate. But we've been through too many technology cycles to confuse a compelling demonstration with a proven business outcome.
The dairy industry doesn’t need AI to invent a new version of the truth. It needs AI to help people make better use of the truth their businesses already hold.