Measuring potential: inside 400 video analyses of young sport horses
Between May and October we analysed close to 400 horses for our pilots, mostly dressage prospects with around 50 jumpers, and most of them too young to be ridden. Here is how we did it, what the data could and could not tell us, and why the one question everyone asked is also the hardest one to answer honestly.
The question every pilot asked
It arrived in different words, but it was always the same question. A breeder sent a clip of a yearling loose in the arena. A dealer sent a sales video. A trainer sent four phone clips of a three-year-old along the wall of an indoor school. And then: "Is this one a Grand Prix prospect?"
Close to four hundred times since May, we tried to answer. Not with a yes or a no, because nobody can promise a Grand Prix career from a video of an unbacked horse, but with measurements: how far the hind foot lands beyond the front print, how the topline swings in each gait, how the neck and pelvis move, what the shoulder and croup angles are, and how the horse carries its tension. This article is about that process: what we built, what worked, and what did not.
Who we are
Equinnovate is an AI company that brings measurement to equestrian sport. Our focus is sport horses. We build models that read a horse from images and video and we wrap them in service layers for the people who make decisions about horses and the sport:
Riders who want to know what their horse's movement and temperament look like in numbers, and how they change with training.
Trainers who need objective markers of progress.
Breeders who want to compare foals and young horses on something more repeatable than an eye on a wet Tuesday.
Judges who want their impressions backed by measurable indicators.
The equestrian world runs on experience and intuition, and we have no intention of replacing either. What we add is a ruler: the same measurement, applied the same way, to every horse.
How we ran 400 analyses in six months
Every horse went through the same process, whatever the quality of the footage. Around 50 of the 400 were jumpers; the rest were dressage prospects, and dressage is the lens of this article.
Our internal pose models read the horse from the video. On top of that sit the rules we have built with riders, trainers and judges over the past two years: they turn the raw tracking into the things an experienced eye looks for, in three families. Gait quality for the work the horse is intended for, with dressage as the main lens. Emotional state, because a young horse shown loose in a strange arena is rarely calm, and the analysis has to tell a tense horse from a tense moment. And morphometry, where the footage allows it, with every proportion and angle carrying a confidence level, because a trotting horse is not a square standing photo.
Our AI read the papers too, and checked whether what the pedigree is known for actually showed up in the video.
Every horse left with a written assessment a person can act on: a summary, the measurements we trust, the things we could not see, and what to film or check next. We would rather write "not measurable from this footage" than print a number we do not stand behind.
What we learned about young, unridden horses
Most of the 400 were not of riding age: foals beside their dams, yearlings, two- and three-year-olds shown loose or in hand. That shaped everything.
The walk is the most honest gait, and the least filmed. Sales videos love the trot. But the walk is the gait that training changes least, and its faults are the hardest to fix later. It is also the gait we can measure most reliably from an ordinary phone clip. Yet in many videos the walk lasted three seconds, filmed while the camera was still finding the horse.
A loose young horse shows you its ceiling and its temperament, not its average. Chased round an arena, a two-year-old gives you its most spectacular trot and its most tense neck in the same ten seconds. Learning to read the two separately was one of the main things the 400 taught us.
Expression is easy to see. What matters is harder. The flashy front leg is what sells a video. The qualities that carry a horse up the levels are quieter, and they are the ones we spent most of our effort measuring.
Conformation from video is possible, with honesty. Some proportions and angles can be measured well from the right frames; others cannot, and we said so each time.
Pedigree explains, it does not predict. When a sire is known for a quality and the video shows it, that is a useful confirmation. When a fashionable young sire has no sport-age offspring yet, the pedigree is a hope, not a data point, and we wrote it that way.
Why a video with a rider changes everything
A minority of pilots sent footage of the horse under saddle, usually a three- or four-year-old a few months into work. Those analyses were in a different league.
A rider sets the tempo and the line, so we get repeated, comparable strides instead of a horse galloping past the camera once. We see the horse under a load, which is the only condition in which the questions that matter for a dressage career can be answered. And the horse's emotional state becomes readable: tension in a loose horse is mostly noise, tension under a quiet rider is signal.
If we could change one thing about the 400, it would be this: more ridden footage, even rough, even on a phone, even of a four-year-old who has been backed for six weeks.
Grand Prix prospect estimation: the question we were asked most
No other request came close. Breeders, dealers and trainers alike wanted one thing from the analysis: an estimate of whether this horse could one day do Grand Prix.
We took the question seriously, and we refused to answer it with a single score.
Grand Prix is collection. The physical and mental ingredients that make collection possible are visible early, and most of them can be measured from the right footage. That is what our Grand Prix prospect profile does: it sets each horse against those ingredients, with measured values where we have them and open questions where we do not.
What we do not pretend to know is just as important: soundness, how the horse will take the contact, the quality of the training it will get, and the plain fact that most horses with the right ingredients never reach Grand Prix because something else happens first. Every report says so.
So the conclusion reads like "two of the three gaits are already Grand Prix-type; the hind leg and relaxation under saddle are the open questions," not "82% Grand Prix probability." Pilots told us this was more useful, because it tells them what to look at next, what to film next, and what to ask the vet.
The honest finding from 400 horses: spectacular trots were common. The full combination of ingredients was not, and it almost never came from the horse with the flashiest sales video.
When the X-rays arrive
For part of the 400, pilots also shared the veterinary file: pre-purchase radiographs, sometimes a full set, sometimes just the findings that had made a buyer walk away.
Three words came up again and again: kissing spines, ECVM and OCD. Each of them scares people, and each of them ended a negotiation more than once in the footage we saw. Yet all three are common findings in young sport horses, and a finding on a radiograph is not the same thing as a horse that will never do anything.
Overriding dorsal spinous processes show up on the X-rays of many horses that are working comfortably at every level. ECVM, a congenital variation of the lower neck vertebrae, is found in horses competing at the top of the sport and in horses that never had a problem; what it means for an individual is still debated among veterinarians. OCD lesions are frequent in young warmbloods, and depending on where they sit and how they are managed, many of them never affect a career.
None of this means a finding should be ignored. It means the finding is one variable, and on its own it is a poor basis for a decision in either direction. The question a buyer actually has is not "does this horse have kissing spines?" but "does this horse, with this back, this topline, this way of moving and this temperament, have a realistic path to the work I want from it?" The vet answers the first question. Nobody has been answering the second one with data.
That is what we are working on now: putting radiographic findings next to the movement and conformation measurements of the same horse, so that selection and de-selection take more variables into account than one worrying image. A horse with a radiographic finding and a back that swings freely under a rider is a different proposition from a horse with the same finding and a tight, held topline. Measuring that difference is the point.
We are not veterinarians, and the clinical judgement stays with them. What we can add is context, and context is what is missing when a buyer looks at one X-ray and walks away from a horse that might have gone all the way.
How to film for a useful analysis
Most of what limited our analyses was footage, not models. Three things make the biggest difference: a long, calm side-on walk; trot and canter on both reins filmed from a fixed position rather than chased with the camera; and ridden footage as soon as the horse is old enough. A square standing photo and the papers complete the picture. Pilots receive our full filming guide with their first analysis.
What's next
Six months and 400 analyses gave us two things: a calibrated set of rules that we trust more than we did at the start, and a clear view of the gap between what pilots want to know and what a loose-horse video can tell them. The next steps follow from that: better pose models that hold up on tense, fast-moving youngsters, ridden-work analyses that measure what matters once the horse carries a rider, and the work of bringing veterinary findings into the same picture as movement and conformation.
If you breed, train, sell or judge sport horses and want numbers next to your eye, get in touch. Send us a video. We will tell you what we can measure, and what we cannot.