Au2mate works with food and beverage producers to turn production data into decisions that move operations forward. What we consistently find is that the most common barrier is not a lack of data. It is a lack of confidence in it.

Henrik Kæmpe, Operations Development Manager at Mammen Mejerierne, one of Denmark’s largest dairies, describes the problem directly. When operators cannot trust what they see, they do not act on it. For a facility logging thousands of data points every day, that is a significant problem.

The sensor is usually fine. The question is whether anyone believes it.

Most modern dairy facilities have sensors covering temperature, flow, pH, and pressure.

The readings are logged. The systems are running. But operators have learned, through experience, that a sensor can drift out of calibration, a pipeline shift can assign a volume to the wrong batch, or a reading can keep the line moving even after it has stopped being accurate.

Operators who reach this point do not simply ignore the data. They weigh it against what experience tells them the number should be. When the two diverge, instinct wins. This is a rational response to a system they have learned not to fully trust, and it means the data investment is doing only a fraction of the work it should.

Validation is an ongoing task

One of the lessons from Mammen Mejerierne’s data infrastructure work is that validation has to be built into the process continuously, not completed at the start. When the facility began collecting more data than before, one of the first questions they ran into was not what to collect. It was whether what they were collecting was correct.

• What happens to the milk in the pipes when you switch from one batch to the next?
• Does that volume belong to the outgoing batch or the incoming one?
• If a sensor breaks down mid-run, what number gets logged?

These cases are all regular operational decisions that, if handled inconsistently, will erode the reliability of every downstream calculation like yield calculations and traceability records. All of it rests on the inputs being trustworthy.

The KPI that nobody owns is the KPI that nobody fixes

To Kæmpe, data trust is also an organisational challenge. He describes a pattern familiar to anyone who has worked with production dashboards: A KPI turns yellow, and the people looking at it cannot trace it back to decisions they make each day. They keep watching it, because it is on the board but nothing changes.

The fix here is rarely more data.

It is making sure the right data is visible to the right people, at the right level of the organisation, connected to decisions they are in a position to make. A manager should not spend time on a KPI that operators can and should resolve. When those boundaries blur, people stop trusting the whole system.

Trust is built deliberately.

The path from raw sensor output to confident, consistent decision-making does not happen automatically.

It requires the people closest to the production floor to be involved in deciding which data matters to them, what it means when a number goes out of range, and what action is expected.

Kæmpe describes the satisfaction when the system is working properly: sitting in on a team meeting, hearing operators and a team lead work through a KPI outlier, and recognising that the infrastructure built to support those conversations is actually being used, and that the data reached the right person, at the right level, in a form they could act on.

That is the goal. Data that lands.

The technology to capture production data from a dairy facility is widely available, but the hard work is making that data something the organisation can rely on.