Every activity-based heat detection system in widespread use today is built on a shared assumption: that cows cycle on approximately a 21-day schedule, and that elevated accelerometer readings around that interval signal oestrus. The assumption is biologically sound at the population level. The problem is that individual cows deviate significantly and consistently from the 21-day average, and when a system doesn't account for individual deviation, it introduces timing errors that reduce detection accuracy.
This article explains the physiology behind cycle length variation, quantifies how much individual cows deviate from population norms, and describes why building a cycle history per cow, rather than relying on a species-level average, produces materially more accurate detection outputs.
What "21-Day Cycle" Actually Means in Practice
The bovine oestrous cycle runs from the onset of one oestrus event to the onset of the next. The average across the Holstein and Friesian dairy cattle population is commonly cited at 21 days, with most published ranges spanning 18 to 24 days. This population range is correct, but it describes variance across animals, not within a single animal over time.
Within-cow cycle length is more consistent than between-cow variation, but it is not perfectly constant. Factors including body condition score at different points in the lactation cycle, energy balance in early lactation, heat stress events, and luteal phase length all introduce variation. A cow that averaged 20-day cycles in her first lactation may shift to 22-day cycles in her third if her early lactation energy balance has changed.
The practical consequence: if a detection system is looking for an activity peak around day 21 from the last detected heat, it will be looking 1 to 3 days early for a cow whose natural cycle runs 22 to 24 days. That cow's activity peak will occur outside the detection window, and the heat event may be flagged as low-confidence or missed entirely.
The Magnitude of Individual Deviation
Published reproductive physiology literature on Holstein and Friesian cattle reports inter-cow standard deviations in cycle length of approximately 2 to 3 days from the population mean. This seems small until you consider how it translates into detection timing.
A system calibrated to a 21-day window with a plus-or-minus 2-day detection buffer covers cycle lengths from 19 to 23 days. A cow who consistently cycles at 24 days sits outside that buffer. Her third cycle event falls on day 24, not day 21, but the detection system's confidence is highest around day 21. If activity data on day 21 is ambiguous, the system may flag a false negative event for a heat that's actually going to occur in three days.
In a herd where 15 to 20% of animals have consistent cycle lengths outside the 19 to 23 day range, this matters. These cows are not reproductively abnormal. They simply have personal baseline cycles that sit in the tails of the population distribution. Treating them identically to the population mean introduces a systematic detection error for that sub-population of animals on every cycle, not just occasionally.
What Cycle History Actually Captures
A per-cow cycle history records the intervals between detected heat events across multiple cycles. After two confirmed events, a system can calculate the cow's own average interval. After three or four confirmed events, it can begin to estimate whether the cow has a stable personal cycle length, a slow trend in either direction, or high cycle-to-cycle variability.
The detection benefit is direct. Instead of flagging elevated activity around "day 21 from last event", the system flags elevated activity around "day 22.5, based on this cow's last two cycles" or "day 20, because this cow's history runs short". The detection window narrows to a more precise expected range, reducing both false negatives (missed heats that occur outside the population window) and false positives (ambiguous activity events that look like oestrus but are timed wrongly).
This is the core distinction between population-level detection and individual-level detection. Both use activity collar data. Only individual-level detection treats cycle timing as a variable that must be personalised, not assumed.
Where the Improvement Shows Up
The clearest gain from cycle history appears in two cow categories. First, animals with consistently short or long cycles relative to the population mean. These cows are missed more often by population-calibrated systems, not because their heat expression is weak, but because the detection window is miscalibrated for them specifically. Cycle history corrects the window.
Second, cows in the early post-partum period, where cycle length is often longer and more variable as the reproductive system re-establishes normal luteal function after calving. The first two to three cycles post-calving frequently run longer than the cow's steady-state interval. A system without cycle history has no way to distinguish a cow in her post-partum resumption from a cow who simply has a long natural cycle. Both are at risk of missed detection. Cycle history at least narrows the uncertainty for cows with an established record.
We want to be honest about what cycle history doesn't fix. For a cow with genuinely suppressed heat expression due to energy deficit, lameness, or heat stress, a more precisely timed detection window still may not generate a detectable activity signal. Per-cow cycle history improves timing accuracy. It doesn't manufacture a signal where the underlying oestrous expression is weak. Those cases require attention to the animal's condition, not just the detection algorithm.
The Minimum Record Required
A reasonable question for farm managers is how many cycles are needed before cycle history becomes useful. Our experience is that two confirmed cycle intervals are sufficient to narrow the detection window meaningfully for most cows. A single confirmed interval tells you the cow's last observed cycle length. Two confirmed intervals, particularly if they agree closely, give reasonable confidence that you're looking at a stable personal baseline rather than a noisy one-off measurement.
Three or more intervals allow a more robust estimate and start to reveal whether the cow's cycle length is stable or has a directional trend. For a cow entering her third lactation with a full prior season's cycle record, the detection system is working with substantially more information than it has for a first-calf heifer joining the detection system for the first time.
This is a practical argument for maintaining continuous records rather than starting fresh each season. The value of cycle history compounds over time. A cow in her third year with Ovum installed has two to three years of cycle data informing her detection windows. A cow in her first year has, at best, a partial-season record. The detection accuracy for long-standing cows in the herd is higher than for new entrants, and that difference grows as records accumulate.
Population Models vs Individual Records: A Practical Distinction
Activity collar manufacturers typically publish detection rates based on validation studies conducted on research herds or pilot farms. These rates are real, but they're population-level averages. They tell you that across the herd, a given proportion of heat events were correctly flagged. What they don't tell you is which cows were detected reliably and which were not.
In most herds with population-calibrated detection, the well-detected cows are those whose natural cycle length sits close to the population mean. Cows at the tails of the cycle length distribution, the 20-to-25% whose natural intervals run consistently shorter or longer, are systematically under-detected. The reported detection rate reflects the performance on the easy majority, not on the harder sub-population where missed heats are most likely.
Cycle history directly targets that harder sub-population. It narrows the gap between the detection performance on easy-to-detect cows and the detection performance on cows whose biology diverges from the population average. The result isn't a dramatic headline number. It's a reduction in the proportion of cows who are reliably missed season after season, which is where the real herd-level cost sits.
