Skip to main content

Peer benchmarking

Useful comparison, without exposing anybody

A benchmark is only worth having if the farms in it are genuinely comparable, and only worth participating in if it cannot be reverse-engineered back to a named producer. Both of those are engineering problems, and we treated them as such.

How a cohort is built

Like for like, before anything is averaged

Comparing a 30-week caged Hy-Line W-36 flock in January against a 70-week cage-free brown flock in July tells you nothing. Cohorts are filtered first.

Genetics

Breeder strain, resolved through an alias list so “H&N” and “H&N Nick Chick” land in the same cohort.

Flock age

Grouped into age bands so a week-34 flock is compared against flocks at a comparable stage.

Production type

Caged, enriched, cage-free, free-range and pasture operations are not pooled together.

Season

Environmental load differs enough through the year to matter.

Age banding, season definition, cohort weighting and outlier handling are being finalised with our nutrition team. They are configuration rather than assumptions baked into code, and every benchmark records the definition that produced it — so a report generated today still means the same thing in two years.

Privacy by construction

Four rules that make the benchmark safe to join

A minimum of ten eligible flocks

Below that, the peer figures are suppressed and the screen explains why rather than showing a thin, misleading number. Genetic standards stay visible, so the comparison is never empty.

The threshold is re-checked after every filter

Otherwise a cohort could be narrowed one filter at a time until only one recognisable farm remained. Each additional filter is re-evaluated against the threshold, not just the first.

Consent belongs to the company, and is versioned

Not to a user, and not to a farm profile. It records who decided, when, and against which version of the terms — and it can be withdrawn.

No named peers, ever

Producers see cohort statistics: average, median, a direction-aware top-performer figure, and the number of flocks behind them. Never an individual line belonging to somebody else.

Direction matters

“Top performer” has to mean best, not highest

A naive ninetieth percentile rewards the highest number. For half the metrics on a layer farm, the highest number is the worst outcome.

How the top-performer figure is calculated per metric
Metric Better is
Hen-day production Higher
Hen-housed production Higher
Weekly mortality Lower
Cumulative mortality Lower
Feed per dozen Lower, within a viable range
Average egg weight A target band, not an extreme

Egg weight is the interesting case: neither the smallest nor the largest eggs are the goal, so a percentile is the wrong tool entirely. Metrics like that are handled as a target band, and the rule for each one is set by our nutrition team rather than inferred from the data.

Common questions

What producers ask before opting in

Can another producer work out my numbers?
No. Peer results are only ever cohort statistics, cohorts are suppressed below ten eligible flocks, and the threshold is re-applied after every filter so a cohort cannot be narrowed down to an individual. Benchmark queries are also logged, so repeated narrowing attempts are detectable after the fact.
What happens if I do not opt in?
You keep full access to your own dashboards, history, alerts, feed analysis and reports, and you can still compare against genetic standards. You simply do not contribute to, or receive, peer figures.
Can I change my mind?
Yes. Consent is recorded at company level with a version, a date and the person who decided, and it can be withdrawn. Withdrawal takes effect on the next benchmark computation.
Do Devenish staff see my identifiable data?
A nutritionist sees the customers assigned to them, because that is their job. Cross-customer analytics are built from authorised data only, and internal access to sensitive exports is audited.
Why is my benchmark sometimes unavailable?
Because the cohort you filtered to had fewer than ten eligible flocks. Widening the filters usually restores it. We would rather show nothing than show a number derived from three farms.