Mastitis detection remains constrained by parlor realities. Modern dairies are designed to maximize throughput, leaving little margin for detailed milk inspection on every cow at every milking. Even highly trained milkers can overlook subtle milk changes or early signs of disease when operating under fatigue, time pressure and competing demands.
“With how fast parlors are being pushed, workers are asked to milk more cows in shorter amounts of time. To look at and examine milk thoroughly for 8- or 12-hour shifts, it doesn’t always happen on every single cow,” says Dr. Justin Hess of Clinton Veterinary Services. “You’d be amazed at how much you can actually miss.”
Subclinical mastitis is particularly vulnerable to underdetection because it requires intentional testing that is accompanied by labor, cost and workflow implications.
Improving mastitis outcomes depends less on detection itself and more on what happens afterward. Farms today are generating more information than ever, but that information does not automatically translate into better decisions. Sound mastitis protocols need to be in place and understood by all on a dairy.
“If you try to develop a protocol, and the management team isn’t on board and you don’t have the right people in place, you’re going to struggle and probably make things more difficult,” Hess explains. “We like to keep things simple but effective.”
These protocols largely include management choices surrounding animal density, mastitis detection methods and even the choice of bedding in the stalls.
Concerning mastitis detection methods, on-farm culturing demonstrates the tension between simple and complex protocols well.
“Culturing on-farm can be a struggle because of the increase in labor and having a dedicated person to do it. You also need the knowledge and desire to do it and do it correctly,” Hess says.
When farms have dedicated personnel, clear interpretation guidelines and confidence in how results will be used, culturing can reduce unnecessary antibiotic use and improve outcomes. When those conditions are absent, culturing may delay treatment without changing behavior, prompting farms to revert to broad-spectrum approaches for the sake of speed.
The challenge isn’t just the size of the farm, but the speed at which data must be converted into a treatment decision.
As the limitations of manual culturing and visual inspection become more apparent, the industry is shifting toward passive detection — systems that monitor the cow without requiring extra labor hours.
To address the complexity of dairy systems, Dr. Alon Arazi, chief veterinarian at Afimilk, hopes consolidating data generated by monitoring animals in existing protocols will help refine management and improve animal health.
“All this data is being gathered into one piece of software in which we do the analysis to detect mastitis,” Arazi says.
Sensor systems can also be used to detect mastitis based on deviations from the norm at a cow level. This baseline varies for each cow, meaning you need historical data for comparison.
“The main way to detect mastitis is based on what’s normal [for that animal]. Increased conductivity of a cow or dropped lactose to a lower level than is expected. This is mainly happening with clinical mastitis,” Arazi says. “One of the problems with subclinical mastitis is that the changes sometimes are very, very low and very hard to detect. In that case, we are looking for more and more sophisticated modeling algorithms that combine more and more things together to see things that are just starting to change.”
Mastitis Indicators Used in Automated Monitoring Systems
Automated monitoring systems identify cows suspected of mastitis by analyzing multiple milk and cow-level parameters simultaneously, rather than relying on a single signal. Key indicators include:
- Milk conductivity
- Increased electrical conductivity associated with changes in ion flow during mastitis
- One of the primary and earliest milk signals used
- Milk yield
- Sudden or unexpected drops in production relative to the cow’s baseline
- Lactose concentration
- Decreases in lactose production when udder function is impaired
- Possible lactose leakage from milk or utilization by bacteria
- Milk flow / milking dynamics
- Changes in milk flow rate that may reflect udder discomfort or inflammation
- Rumination patterns
- Decreases in rumination associated with illness or discomfort
- Eating behavior / dry matter intake
- Reduced intake relative to expected performance
- Activity and behavior changes
- Deviations from individual cow behavioral baselines
This collected data is then compared and put into context on the individual, group and herd levels. Mastitis alerts are generated by combining multiple indicators, rather than any single threshold.
These disparate data points, along with the sheer volume of data, are where machine learning thrives.
“AI or machine learning will allow you to detect things that, even for us, are hard to see now. This for sure will improve subclinical detection,” Arazi predicts.
These systems aim to provide directional insight that shortens the time between detection and action by reducing the workload and finding changes in cow performance before they would be noticed by a worker. Catching a case 24 hours earlier could be the difference between a quick recovery and a culled cow.
“You don’t have to check every cow because the system has checked every cow two or three times in a day depending on how many milkings there are,” Arazi says. “You get the information, and you get the option to catch things earlier than people can see with their eyes.”
The Human Filter: Why Detection Requires Interpretation
Alerts without context quickly become noise. High alert frequency, poor specificity or unclear next steps can erode trust in the system. This is where veterinary intervention can help a dairy understand what they’re seeing and how best to act.
Hess stressed the questions he poses to dairies implementing updated mastitis detection protocols: “When you have that information, what are you going to do with that information? Are you going to actually change your protocols?”
Having more data is only useful for improving animal management if accompanied by a plan to act on what that data is telling you.
Technologies offering continuous observation and reduced reliance on human detection can introduce risks related to accuracy, workflow fit and trust. There is also the worry of false alerts.
“We can still improve accuracy, reduce false alerts and get more sensitivity,” Arazi says, speaking on the Afimilk system for mastitis detection.
These systems are, of course, not infallible. As with all hardware, there are uncontrollable hiccups that need to be considered when looking at the data generated.
“There are some critical parts of measuring conductivity,” Hess says. “If milk is moving or if air gets into the system, it can affect the sensitivity or the reading on it.”
At their core, these tools are designed to flag abnormal patterns, not to dictate diagnoses or management decisions. Alerts of deviations are only meaningful after interpretation by people who understand the cows, the parlor and the operation of the farm. Without the human layer, accurate detection risks becoming background noise.
“The only thing worse than no data is having wrong or misleading data,” Hess says.
The limitation is not simply technological, but decisional. This becomes most apparent when detection systems skew too far toward sensitivity at the expense of specificity. Highly sensitive tools identify earlier or more subtle changes, but they also generate more false positives. Each unnecessary alert pulls time and attention away from other priorities.
At the other end of the spectrum, overly specific systems may miss early disease signals, limiting their preventative value. Effective mastitis detection depends on deliberate trade-offs, favoring actionable accuracy over alert volume. The future of the dairy isn’t just in the data collecting sensors, but in how the person in the office uses that data to provide better care for the cow.
Having spent their careers at the intersection of veterinary medicine and dairy technology, Dr. Hess and Dr. Arazi share a common passion for evolving how we look at herd health. On the first episode of The Bovine Vet Podcast, they join host Andrea Bedford to discuss why mastitis is much more than a simple infection.
