The 2025 breeze up sales season thundered home in record-breaking form. At Tattersalls Ireland, turnover passed the €10 million mark for the first time in the sales’ history.
Across Europe, year-on-year gains confirmed a banner season, with Craven up 25.7 percent, Arqana rising by 24.8 percent, and Goffs UK posting a 12.4 percent increase. New benchmarks were set at the Craven sale on consecutive days: a £1.4 million Acclamation colt bought by Godolphin and a £1.75 million Havana Grey colt snapped up by Amo Racing. The message was unmistakable: the market for elite two-year-olds is not only alive, it’s accelerating.
But beneath the million-pound bids and the buzz of bloodstock agents, a quieter shift is underway. Artificial intelligence is taking its first real stride into the parade ring and quietly shaping the future of breeze up sales, turning gallops into guidance. Stephen Davison, Head of Commercial Operations at Pythia Sports, provided SiGMA News with exclusive insights into the algorithms that catch what even the sharpest eyes might miss.
Inside the data driving breeze up sales
“We run as an advisory service alongside traditional buying,” said Davison. “There’s definitely headroom for it to go further.”
Pythia’s proprietary AI model analyses over 100 stride, biomechanical and time-based factors to generate a performance rating for each horse at the breeze up sales. It doesn’t aim to replace human intuition but to reduce bias and act as a second set of eyes, ones that don’t blink.
“The trainers and bloodstock agents will still pull out horses they like,” Davison explained. “They’ll say: how did lot 73 or 105 run against the data? If our scores match their instincts, they go deeper. If we didn’t like the stride profile or biomechanical walk, and the time was only OK, they might leave it.”
That partnership is key. Rather than unseating tradition, the model is simply riding alongside it.
What the model sees
“Our stride model is about 50-plus individual factors. People say, ‘Is it stride length?’ Is it cadence? It’s many things coupled together,” said Davison. “There’s no magic indicator. But that all plays a part.”
The model, described as a “black box” system, is trained using historical sales data and subsequent race results.
“It looks back at how horses breezed in the past, compares those stride scores, and learns from whether we were right or wrong,” Davison added. “We may upgrade or downgrade based on what actually happened on the track.”
Of course, even the sharpest model can’t clear every hurdle. What it doesn’t see is just as important. “It has no vetting data,” he said. “You might like a horse, but if it doesn’t vet clean, it won’t sell for a big price. Sometimes you’re trying to find value in those gaps.”
The one that got away
While Davison did not name specific horses that the model might have flagged, he acknowledged the allure of hindsight.
“Everyone in racing has a story about a horse that got away. I won’t name names because it wouldn’t be fair,” he said. “But we’ve seen quite a few hit their stride in the last two or three weeks that we had high up on our list. One horse won last night. We didn’t buy it, but it won quite nicely. It’ll probably be targeting Royal Ascot.”
It’s not just the horses Pythia advises on that offer confirmation. “We’ve graded every horse at every sale,” Davison noted. “So, when we see a horse perform that we had high up on our list, even if we didn’t buy it, that’s good confirmation that what we’re doing is working.” And as performance data becomes part of the sales story, so too does the question of care. For a deeper view on how ethics and aftercare are strengthening, explore the recent SiGMA News report on the future of horse welfare in British racing.
Data in the decision-making mix
The ripple effects of AI-led horse assessments aren’t just limited to the sale ring. Bloodstock agents, trainers and owners are using the data to inform decisions, especially when opinions diverge.
“Trainer likes it, bloodstock agent likes it, data likes it, that’s the trifecta. But if one disagrees, the data can help bring clarity. We’re not promising Group 1 glory or overnight miracles. We’re trying to reduce the risk of buying a horse that might not be as good as it looks,” said Davison.
Some trainers and agents revisit Pythia’s ratings once their horses begin racing. “I know they’re checking how we rated the horses now they’re on the track. A couple of bloodstock agents have already been in touch, saying, ‘well done on that one’, or asking why we rated another the way we did,” Davison added. For now, Pythia remains focused on pre-sale analysis, but the possibilities don’t end at the hammer.
From paddock to punter potential
While Pythia’s technology wasn’t designed for betting markets, Davison acknowledged that some bookmakers are already watching.
“One UK bookmaker looks at the breeze up times,” he said. “With two-year-old markets, there’s no form to go off. So, pedigrees, sales prices, and breeze data are the early indicators. Our data could help form early views, but we haven’t had direct engagement yet.”
Real-time betting analytics may still be at the starting gate, but the potential is picking up pace. SiGMA News recently featured how data and disruption are redrawing the lines in UK and US racing. And what about race-day analytics? “The model is designed for track performance and bloodstock, not betting. But I do think it’s something to watch, especially for early two-year-old races.”
Scaling up smart: How far can Pythia’s model run?
With the breeze ups wrapped, Pythia is looking ahead.
“We’ve had people ask about yearlings,” Davison said. “There’s less data, but we’ve got the biomechanics model ready to evaluate walks. We’ll need to adapt it a bit, but we’re definitely looking at it.”
They’re also considering expanding to horses in training sales, where more race data is available. But Davison stressed caution, “Everything we do is about timing and quality control. We won’t rush anything out.”
Could the model ever include health or physiological data? “I’m really not sure. You’ll always need a vet to feel a horse’s legs,” he said.
As for international growth: “We could lift these models and put them into any breeze up around the world. It’s all about scalability.”
Backing brains and bloodlines
From bloodstock ring to betting markets, AI is beginning to influence the way decisions are made. Pythia’s , but in a high-stakes, high-variance world, it offers something increasingly rare: evidence.
In the fast-paced world of breeze up sales, even a half-second insight can change everything. “What matters is how the horse performs on the track,” Davison concluded. “But we’re helping buyers steady their hand before they make the final gallop to the sales ring.”
*Stephen Davison is a commercial operations manager with over 10 years of experience in the sports betting and horse racing industries. A long-time advocate for smarter, data-driven decision-making, he combines a deep love of the sport with a sharp eye for commercial opportunity.
After starting his career at Kambi as a trader, Stephen co-founded Black Type, where he later became Chief Operating Officer (COO). During his time there, he led initiatives across partnerships, operations, and strategic development. Since joining Pythia Sports in 2020, he has focused on building and expanding the company’s B2B offering.
