In a hatchery, the word “AI” currently covers two things that have almost nothing in common. One is a set of systems installed at scale because national law left producers no alternative. The other is a demonstration on a trade show stand. Telling them apart is most of the work.
Key takeaways
- In ovo sexing had reached around 15% of EU layer hens as of 30 September 2023, following national bans on culling day-old male chicks.
- Competing technologies act on different days of incubation, from day 8 to day 13, which is what decides whether one fits your process.
- Camera systems on the chick line quote high sorting accuracy at stated throughputs, always under the supplier’s conditions.
- The dataset most hatcheries under-use is the one they already write down every cycle.
Where the technology is already installed, and why
Layer hatcheries in several European countries no longer have the option of sorting chicks after hatch. Germany, France and Austria have banned the culling of day-old male chicks, and since 2023 French hatcheries have been required to use in ovo sexing technologies, with the French government contributing 10 million euros in subsidies to support the transition. Italy set its own phase-out for 2026.
The adoption figure gives the scale. As of 30 September 2023, in ovo sexing covered roughly 15% of EU layer hens, or 56 million out of 389 million commercial birds, according to figures compiled by AgFunderNews. That is a substantial installed base for a technology that barely existed commercially a decade ago, and it exists because a legal deadline made the business case irrelevant.
The day of incubation is the real specification
Competing in ovo sexing systems are separated less by accuracy claims than by when in incubation they can work, because that determines where the equipment sits in your process and what it does to your transfer schedule. Hyperspectral imaging systems, such as the CHEGGY developed by Agri Advanced Technologies, operate between day 10 and day 13 and are quoted at 20,000 eggs per hour. Allantoic fluid sampling, used by In Ovo, Respeggt, PLANTegg and Seleggt, works at around day 9. Spectroscopy-based approaches from HatchTech and Respeggt, and volatile compound detection from SensIT, target day 8. Orbem’s approach combines magnetic resonance imaging with machine learning.
Throughput is the second axis, and it is where the sector set its own bar: the Egg-Tech Prize used 15,000 eggs per hour as a minimum requirement, which gives a reasonable reference point for judging whether a system can keep up with a commercial setting pattern.
Cameras on the chick line
Downstream of the hatcher, vision systems are being applied to sorting and grading. Reported figures include AI-based sorting of males and females at up to 98% accuracy, and a vent sexing machine quoted at 98.5% accuracy for white and coloured broilers, layers and turkeys at a rate of 1,000 chicks per hour. Optical systems are also used to flag defects such as an open navel.
Two cautions belong with those numbers. Accuracy is quoted under supplier conditions, on the breeds and chick quality the supplier tested. And a sorting rate of 1,000 chicks per hour is a different order of magnitude from a hatch of 100,000, which makes the relevant question where in the process such a system pays for itself rather than whether it works.
| Family | Acts at | Adoption driver | Question to put to the supplier |
|---|---|---|---|
| In ovo sexing | Day 8 to day 13 of incubation | National bans on culling day-old males | What it does to transfer timing and to hatchability |
| Vision-based chick grading | After hatch, on the chick line | Labour cost and uniformity of delivered batches | Accuracy on your breeds, at your line speed |
| Analytics on process data | Across the whole cycle | Data already recorded but not compared | Which decision changes when the alert fires |
The data you already have
The third family is the least photogenic and the most available. Storage duration, source flock, tray position, egg weight loss between set and transfer, breakout results and hatch window are already written down in most hatcheries, and in most hatcheries they are never compared with each other. An analytics layer that joins them answers questions no sensor can: whether a poor hatch belonged to that flock or to that machine, and whether the difference repeats.
Sensors add resolution to that picture rather than replacing it, and the technologies that act on the incubation environment itself are worth assessing on the same terms, as we have done for light, sound and air quality in the hatchery.
How to read a claim before you budget for it
Suppliers in this sector often support a performance figure by pointing at the existence of research rather than at a specific study. A bibliographic database is a search index, not a source, and a claim only becomes checkable once you can name the study behind it.
Four questions usually settle it. Which trial produced the figure, and can you have its reference. How many birds or eggs were involved, and over how many cycles. In which production system and which season, since a result from a single-stage machine in a temperate climate transfers poorly. And how the outcome was measured, because hatchability, hatch of fertile and saleable chicks are three different denominators that a percentage rarely announces.
None of this is scepticism for its own sake. It is the difference between an investment you can defend at the next audit and one you cannot.
Before adding a new data source
Most hatcheries have more usable data than they have questions framed well enough to use it. That is the cheaper place to start.
Sources: AgFunderNews, in ovo sexing penetration in the EU, for the 15% figure as of 30 September 2023 and for the technologies, operating days and throughputs cited; published reporting on national bans on culling day-old male chicks in Germany, France, Austria and on the Italian phase-out; WATTPoultry reporting from the Poultry Tech Summit on chick sex and weight sorting accuracy and throughput. Consulted August 2026. Supplier performance figures are stated for reference conditions and should be confirmed on your own installation.
Published previously, fully revised on 11 August 2026. General technical guidance for hatchery professionals, not an endorsement of any named system.