The sensors work in the papers.They are not in the house.

There is a persistent mismatch in this category between what has been demonstrated and what has been deployed. The demonstrations are genuine. Cameras, microphones and sensor arrays have detected disease, distress and welfare problems in real houses at accuracies that would be useful if a grower had them. Almost none of it is on farms. This note is about the size of that gap, what is actually installed instead, and why we think the sensible response is to build on the second list rather than the first.

Four percent

The cleanest measure of the gap comes from a peer-reviewed review of the poultry precision livestock farming literature. Across 264 publications, a mere 4 percent described commercially available systems. The composition of the rest is informative: 52 percent involved sensors, 42 percent cameras, 14 percent microphones, with 64 percent aimed at health and welfare, 51 percent at production and 8 percent at sustainability. That is a field producing a great deal of capability and very little product.

The same review is blunt about one unresolved piece of hardware. Developing an adequate ammonia sensor, it says, is still a challenge that has not been resolved in a satisfactory way. It also notes that wearable backpack sensors trialled in Flanders substantially interrupted broiler behaviour, which is a reminder that in a house the instrument is part of the system it measures.

What is actually on the farm

The best on-farm number we have found is an MS thesis from the University of Tennessee, and it needs its limits stated before its numbers are used. It surveyed 531 Tennessee broiler producers in January 2024 and got 77 respondents, a 14.5 percent response rate, 95 percent of them contract growers, with 69 percent completion. It is one state, one thesis, not peer reviewed, and it should never be described as a national figure. Non-response in a technology survey plausibly runs toward the technology-interested, which would push the real awareness numbers down rather than up, though that is our inference and not a published finding.

With those caveats, the pattern is stark and consistent. Temperature sensors were reported by 98.1 percent of respondents. Automated feed systems 88.7 percent. Water flow meters 85.2 percent. Humidity sensors 48.2 percent. Then it falls away: video cameras 25.0 percent, data management systems 24.6 percent, automated scales 19.6 percent, automated feed inventory 11.9 percent, location monitors 8.2 percent, gas sensors 7.1 percent, drones 5.1 percent, robots 1.9 percent. Thermal cameras, microphones with an app, and activity monitors were each reported at 0.0 percent.

Awareness, cost, and what growers say would change it

In the same survey, 43.1 percent of respondents had not heard of precision livestock farming and 6.8 percent reported expertise in it. Fifty-seven percent do not use farm-data software of any kind. Asked about barriers, 42.0 percent named cost, 31.0 percent said it does not fit their operation, 27.0 percent complexity and 14.0 percent training.

Asked what would change their decision, 88.4 percent said they would be interested if income were shown to rise, 67 percent said a cost-share would make them more likely, and 61 percent said they need to see it working on a farm first. That last number is the one we take most seriously, because it is not a request for a better spec sheet. It is a request for a reference installation, which is the thing a 4 percent commercialization rate guarantees does not exist for most of this technology.

The researcher side of the same institution surveyed 276 US poultry researchers and got 68 responses. Familiarity with precision livestock farming was 92.4 percent; only 55 percent had or currently used it. Cost was named by 92.9 percent and usability by 87.6 percent. On price, 68.8 percent thought industry would accept something in the zero to twenty thousand dollar band, 23.4 percent twenty to fifty thousand, and 7.8 percent above fifty thousand. That is an opinion survey about willingness to pay, not a quote, and it should not be read as a market price.

The counter-evidence, given fairly

It would be dishonest to write this note as though sensors do not work. They do. In a peer-reviewed 2020 study, an abnormal-sound detection system combining weighted MFCC features with a hidden Markov model ran 24 hours a day for a week in a commercial house of 20,000 white-feather broilers in China and reached 93.8 percent accuracy, 94.4 percent precision, 94.1 percent recall and an F1 of 94.2. That is a commercial house, continuous operation, and a result any grower would take.

The review cited above also reports EU-PLF camera-based early warning at 95 percent true positive events. So the honest framing of the gap is not that the technology fails in the barn. It is that the technology works in the barn and does not reach barns. That is a distribution and economics problem, not a physics problem, and the two have very different solutions.

Sensors work in studies and are not on farms. That is a distribution problem wearing the costume of a technical one.

Deployment failure modes

The catalogue of what goes wrong once hardware is in a house exists, but mostly in an arXiv preprint, which we label as a preprint. It lists dust, ammonia, humidity and lighting interference; sensor and calibration drift; pecking damage; the cleaning burden on cameras; and data lost due to unreliable internet connections. None of that is surprising to anyone who has washed down a house, and all of it is the difference between a device that reads correctly in week one and one that still reads correctly in flock nine. The same preprint records a low-cost MQ-137 ammonia sensor reaching 90 percent similarity to commercial equipment, which is either encouraging or not depending on what you intended to do with the tenth percent.

One thing we will not do is put a number on how many deployments get abandoned. No study we could find reports an abandonment or discontinuation rate for poultry precision livestock farming, and we are not going to invent one. The available evidence is indirect: the gap between 92.4 percent researcher familiarity and 55 percent use, and one documented market withdrawal of a poultry robot. That is worth saying plainly rather than dressing up as a statistic.

A cross-species check

Poultry is not unusual here. In an extension magazine survey of 334 beef producers in North Dakota, South Dakota and Texas in early 2022, a 42.8 percent response rate, adoption of individual precision technologies ran from 0.7 to 18.8 percent: precision mineral supplementation 18.8, daily water monitoring 11.8, breeding monitoring 8.9, virtual fencing 3.6, performance monitoring 3.0. That is cattle, not poultry, and it is outreach material rather than a peer-reviewed study. It is still a useful check that the shape of the problem is not species-specific.

There is one more asymmetry worth naming. The technologies that are near-universal are the ones that were installed for a different reason and happen to produce data as a by-product. A temperature probe is there because the house needs to be ventilated. A water meter is there because water has to be paid for and dosed. Nobody bought either of them as a detection system, which is precisely why they survived a decade of flocks, washdowns and staff changes. Anything sold purely as a detection system has to justify its own maintenance from day one, and that is a much harder position to hold in a poultry house.

The conclusion we draw is narrow. The sensing frontier is not where the value is currently available. The value is in reading the instruments already bolted to the wall, and in giving the grower who said they need to see it working on a farm first something they can actually go and see.