Half your checking interval, on average, and you can work your own figure out in about ten minutes from records already in the drawer. Take the last three problems that cost the operation money, a pump that failed, a mob short at a count, a fuel invoice that stopped matching. Write down the date each one started and the date somebody knew. The average of those differences is your detection delay, and on most operations it is larger than the manager expects and smaller than it needs to be.
The reason the number is rarely measured is not carelessness. Rougoor and colleagues reviewed the research on farm management capacity and reported that no study was found where an effect of the quality of the control itself, as part of the decision-making process, on the farm result could be determined. Pieces of it had been measured, such as time spent on health records and how regularly milkers were spoken to about performance. What the review could not find was work that isolated the quality of the control itself. Control is the function that gets audited least, and a farm that wants the number has to produce it itself.
Why the interval decides the delay before anything else does
Your checking interval sets a floor that no amount of attention can get under. If you look at something every 30 days, and a problem is equally likely to start on any day in between, the average problem has been running for 15 days when you meet it. Half the interval. That arithmetic holds for a water point and for a bulk fuel tank alike.
The floor matters because it is set before any judgment is involved. A sharp manager on a 30-day cycle still averages 15 days of delay. A careless one on a 7-day cycle averages three and a half. Interval beats attentiveness, and the interval is the part a manager fixes by decision rather than by effort, which is why it belongs with Planning and not with the tools that do the checking.
The same arithmetic runs in years wherever the checking is a laboratory analysis rather than a walk. Soil fertility is the plainest case: a declared sampling interval of four years puts the average change two years old by the time a report reveals it, and the fields that are already past that interval can be listed in an afternoon.
What the recording standards already concede
Standards bodies have published their intervals, and reading them as detection delays is uncomfortable. The table below converts each published interval into the average age a problem has reached at discovery.
| What is being checked | Published interval | Average age at discovery |
|---|---|---|
| Lactating dairy cows, welfare inspection | At least once a day | Half a day |
| Field maize, pest scouting before pollination | Every 7 days | 3 to 4 days |
| Dairy herd, milk recording, four-week reference method | 22 to 37 days | 11 to 19 days |
| Northern rangeland property, muster, the round-up when cattle are brought in and handled | About twice a year | Roughly 3 months |
| Dairy cattle generally, welfare inspection | Appropriate to the production system | Not specified |
The last row carries no number at all. The World Organisation for Animal Health sets a daily floor for lactating cows and, for dairy cattle generally, requires inspection at intervals appropriate to the production system and the risks to the health and welfare of the cattle. That is a real requirement with no number in it, which leaves the interval to whoever runs the place.
The dairy recording figures come from the international guidelines for milk recording, where the four-week reference method allows 22 to 37 days between records. At the far end, Meat and Livestock Australia describes northern properties being mustered typically twice each year, and a property mustered twice is not the same as every animal in it being examined twice.
The one document that names the trade-off out loud
Extension guidance for crop scouting states the principle that the livestock standards leave implicit. The University of Wisconsin scouting manual instructs that field visits must be scheduled such that increases in pest populations are detected as soon as economic thresholds are reached, setting weekly visits until pollination and relaxing to roughly ten days afterwards, because by then there is little danger of pest levels passing the economic threshold between visits, the count at which treating pays for itself.
The manual sets its interval against the rate at which the damage accumulates, and loosens it only where it can argue the damage has slowed down. That is the whole method, written down by someone who had to defend it.
Why being there every day is not the same as knowing
Daily contact does not close the gap on its own, and there is a measured example. German researchers compared lameness prevalence assessed by veterinarians against what the farmers themselves estimated across three regions. Vets found median prevalences of 23.1, 39.1 and 23.2 percent. The same farmers estimated 9.5, 9.5 and 7.1 percent. On average, farmers were conscious of only 45.3, 24.0 and 30.0 percent of their lame cows.
The lame cows in that study are animals the farmer walks past every day. The problem is visible, the access is unlimited, and three quarters of it still goes unregistered in the worst region. Presence is not observation, and observation without a written figure to compare against is not detection.
Why the record is always later than the event
The event happens in the field and the record happens in the office, and the distance between them is filled with memory. A 2024 review of on-farm recordkeeping quotes Buckmaster and colleagues to the effect that data quality decreases if it is not recorded in real time, then adds in its own words that infrequent access to the logbook, typically kept in the office, means that data quality heavily relies on the workers’ memory and motivation.
The write-up adds a second delay on top of the first. The problem waits for the check, and the check waits for the write-up. Anyone reconciling records at the end of a cycle is reading a document that was already approximate when it was made.
What a shorter interval actually buys
Shortening the interval improves what a measurement system can detect, and the improvement arrives unevenly rather than in proportion. Researchers building a model to predict subclinical mastitis masked their data to simulate less frequent recording. Two numbers describe how well it worked: how many of the real cases it caught, and how often it left a healthy cow alone instead of raising a false alarm. At weekly recording, which is how the data was actually collected, the model caught 69.45 percent of the cases and correctly left 95.64 percent of the healthy cows alone. Simulating recording only every 60 days, the scores fell to 66.93 and 80.43 percent, and the authors report both degrading as frequency drops.
Two endpoints hide the shape in between, and the middle is where most operations live. In the study’s own table the catch rate does not slide evenly. It bottoms out at 62.11 percent at a 30-day interval before partially recovering, while the false-alarm side gets steadily worse and loses more than fifteen points across the range. A monthly cycle sits at the worst catch rate the study measured, which is worth knowing for anyone who checks monthly because the calendar has months in it. A long interval does not only find problems late. It degrades what it finds, and a manager who stops trusting the alerts stops acting on them.
The mastitis study forecasts onset rather than measuring how long a real case went unnoticed, so treat it as evidence about what a measurement interval can support, not as a detection-delay figure.
Where to start, and what to do with the number
Take the last three problems that cost the operation money. For each, write the date it began and the date it was known, and average the difference. That average is your detection delay, and it took no new system to produce, because both dates already sit in an invoice, a service record or a message somebody sent.
Two comparisons make good first candidates, because both run on records the farm already keeps and both end in a dated finding somebody can be made responsible for. Fuel bought against hours worked produces a gap that has four ordinary explanations before it has an alarming one. The factory service interval produces a number of hours overdue. Neither of them is a diagnosis on its own, and both take a decision to close, which is the part that gets skipped.
The number is worth having because it converts an argument into a decision with an owner. “We should keep a closer eye on things” cannot be assigned to anyone or checked next cycle. “Water points move from a four-day to a two-day check, Marcus owns it, we measure the delay again at the end of the cycle” can be. The four functions of management only close when the last one names a person and a date, and the detection delay is one of the few numbers on a farm a manager can cut by decision alone, without buying anything.