Exclusive panel: JBS Couros shares AI leather grading rollout at APLF

At the Leather Supply Chain Conference ahead of APLF 2026, Ray Connor (COO, Mindhive Global) and Sergio Sfreddo (Head of Global Quality, JBS Couros) sat down with Martin Ricker from International Leather Maker to talk through what it actually looks like to implement AI-powered grading at scale.

This exclusive panel titled ‘The partnership perspective: When global scale meets verified data’ covered why JBS Couros decided to partner with Mindhive Global, what the rollout across 10 Brazilian wet-blue sites actually looked like, and where verified hide data is taking them next.

When grading varies by site, by shift, or by person

JBS Couros runs 21 production units with 7,000 employees across four continents. They're not an operation that lacks process. They'd already built something close to world-class for wet-blue grading. Their problem was grading consistency and hide allocation at scale.

Sergio explained it plainly: "If we think about a larger operation like JBS, we have 10 wet-blue sites in Brazil, two or three shifts depending on the region, and then you have lots of different people working on leather grading. You need to keep consistency and you need to apply the correct hide to the final product. So it's actually a real challenge for us."

This means 20 to 30 people applying what is fundamentally a subjective judgment. What's a TR1, TR2, or TR3? That judgment gets affected by how long someone's been on shift, how they slept the night before, what the market looks like that week.

The downstream consequence: hides pre-selected at wet-blue sites still had to be re-inspected at sorting and finishing facilities before order allocation. Double-handling is costly whether it's the wrong hide sent to the wrong order or over-delivering to customers to avoid claims. Ray put it bluntly: 

"You're sending better quality product to your customers rather than what you actually have to, just to make sure you're not going to get the claims back."

How the partnership started

JBS came to Mindhive through a shared business contact. But what moved things forward, according to Sergio, was alignment on where the industry needs to go. "We got to know Mindhive through a business partner. Through our conversations we could see we share the same view of the future of leather and how to actually achieve something. That's why we see Mindhive as a good partner for JBS."

Ray described the early engagement from a different perspective. The challenge set by JBS's CEO, Guilherme Motta, was clear: understand the business before proposing a solution. "Guilherme said, look Ray, you need to come and understand our business first before you talk to us about what you can do."

JBS opened the doors. Mindhive spent time learning how JBS's grading method had evolved, what made it consistent, what was making consistency hard to hold at the edges. The goal was digitizing something already good, not replacing it with something foreign.

Ray was clear about what they found: "They already had really great systems. They already had a world-class method for selecting different articles at various stages. So it wasn't that we were coming in with a radical new way of doing something because JBS are already a world-class operator."

The challenge was taking that artisanal process and standardizing it across a geographically distributed operating model. And that’s where Mindhive’s global expertise came in. Encoding JBS's own grading logic into a system that could run consistently across 13 sites, while accounting for the regional defect variation between north and south Brazil, without losing what made the method good in the first place.

JBS ran a pilot at their Cactus facility in the US for close to a year. Different raw material. Different conditions. Deliberate. That pilot gave them confidence in both the technology and how Mindhive operated as a partner.

What the rollout actually looked like

The first Brazilian site went live with Mindhive BlueSelect™ in early 2025. Getting the system properly calibrated for Brazilian hides took four to five months. The defect catalog from North American grading had around 16 to 20 classes. Brazilian hides needed more than 30. "It was quite an interesting process to see how we can teach the system and how we need to do that, and our team was very helpful with that," Sergio said.

And in less than 12 months from that first site, Mindhive rolled out systems across the remaining nine Brazilian wet-blue facilities. The team encountered regional variation within Brazil itself. Hides from the north are meaningfully different from hides in the south, which meant each site needed adjustment. "We started in January last year and we finished all 10 sites in December. So it took a whole year to be able to put this system working fine in all the sites."

Martin asked about adoption. Sergio admitted he expected resistance from graders. It didn't materialize. "Actually I thought we would have more resistance but that didn't happen. We kind of embraced everyone since the beginning."

The explanation was practical. Graders weren't removed from the process. Their role shifted.

"Before we had the inspector checking the hides and then we had supervisors and people checking their work. What happens now is that the machine is doing the work allowing the graders to extract and monitor the data. So they transformed from being checked to the one that's actually operating the system and seeing if the system is working fine."

Where the data goes next

The first phase was getting grading right consistently across all 10 sites. JBS is nearly done with that.
Phase two is what the data makes possible. "Phase two is having the system cross-checking our inventory against our orders and helping us to apply the best hide to each product, which is the final objective for the project," Sergio explained.

Three Mindhive BlueSort™ installations are now rolling out across JBS's Brazilian operations. BlueSort™ is an intelligent order fulfillment software solution that uses verified quality data from BlueSelect™ to optimize how inventory is allocated to customer requirements. The aim: reduce lead times, eliminate re-inspection, cut the rate of wrong-hide errors that currently represent real and recurring cost.

Ray described what else becomes possible: "We're able to analyze certain impacts from different regions, seasonal impacts across grades. We're able to look at variability in different types of defects and understand what linkages they have between different types of grades. We're able to help JBS figure out how they can refine and adjust some of the criteria for selection to finely tune the rules that they've built."

This is the shift from rules set once to rules refined continuously. Ray put it another way: 

"Once you've got data, you can't really put the genie back in the bottle. Now that we've got data, what we're able to do with that data really changes the way you can look at the operation."

Questions from the room

Thomas Ferentzi from Bader Group asked which defects are hardest to detect across different stages.

Ray's answer was pragmatic: "The one that matters the most." Different customers have different priorities. Skin diseases matter hugely for some. Butchers' cuts or holes matter more for others depending on intended use. Sergio added honest detail about where work remains: "We still have a little work to do in grubs. Minimal skin diseases are very good results. Open defects, bites, everything, the system is very good."

The follow-up was about looseness and firm grain, one of the industry's persistent challenges. Ray explained how machine vision detects what typically requires tactile testing: "You can see subtle differences in the grain structure and the pattern. Just as somebody who's got 20 or 30 years of experience will be able to tangibly see the difference, we're able to use that methodology in machine vision." Sergio acknowledged the challenge ahead, particularly for automotive leather where looseness is harder to detect visually than in furniture hides.

Matthieu Vassel from Bureau Veritas asked how they keep data attached to individual hides through to cutting, especially through splitting.

Sergio explained their current approach: pallet-level sequencing with unique codes. "We have a unique code for each pallet and the system knows the sequence of the hide inside the pallet. So we can find a specific hide inside the inventory." Moving to individual hide-level traceability with barcodes or QR codes is the next step.

Ray's response focused on pragmatism. Individual hide laser marking is technically possible. Mindhive can do it today. But does the value justify the investment? "If EUDR was more than just batch-based, we'd be able to deliver full traceability through the tannery. The question is who gets the value and who ultimately pays. What we try to do is give a healthy dose of pragmatism and say, what's the best solution that delivers the outcome they need without over-engineering it. The last thing you want to do is over-engineer because that just adds cost for no real value."

Advice for others considering this path

Ray described what verified hide data fundamentally shifts: "Historically, when you've got hides on a pallet and those hides are graded, you either have to manually go through that process of re-selection or you're just leaving too much value there. It's a hard enough industry. We're under pressure from every regard. We've got different types of materials competing against us. Historical prices that we've been able to achieve, they're just not there in the same way that they were. We've got to treat every dollar as a prisoner and fight harder to make sure we can remain profitable."

The way to do that is unlocking best process and practice, but also using information and data differently to make informed decisions.vSergio's view was equally clear: "We see innovation as an important part of the business but we understand it takes time. We are now working on understanding how to use the data that we didn't have before. The final goal is to bring value to the whole chain, not only to JBS but to our customers mainly."

For other tanneries considering this path, Ray's advice was direct:

"The sooner you start, the sooner you're going to realize the value. I think sitting on the fence and thinking 'how does technology apply?' is one approach, but technology is changing in every regard. Even in traditional industries such as this one, the sooner we embrace and adopt the new tools and new ways of working, the sooner we can unlock the value."

But he emphasized the design phase matters more than speed: "We started with a foundation of data first. We built the rules or the criteria well ahead of actually putting the first system in. We spent a lot of time designing the solution and getting the foundation and the core right."