A Kansas Beef Plant Cut Wrong-Order Complaints by 75%. It Started Where the Box Gets Closed.

THE CRAFT

Where, exactly, does a customer complaint begin in your business? A Kansas beef plant went looking for that answer, and found most of it on one line.

Most companies can tell you where complaints end up. They land in customer service, in a credit memo, in a claims log somebody reviews on Monday. Far fewer can point to the spot on the floor, or the step in the process, where the mistake that caused the complaint was first made. The complaint shows up days later and miles away, and by then it looks like a customer service problem.

The plant belongs to Creekstone Farms, which runs a single large beef plant in Arkansas City, Kansas, processing Black Angus cattle for steakhouses, grocers, and export customers in dozens of countries. Its chief operating officer, Andrew Baltz, went after one moment in particular: pack-off, where a cut of beef is bagged, boxed, weighed, labeled, and sent toward the customer who ordered it. In September 2025 the plant finished putting AI vision and automated checks on that line.

“Since fully implementing [the] technology in September 2025, customer claims for order issues have decreased 75%,” Baltz said in the company’s announcement. The goal, in his words, was to “eliminate unintentional oversight, allowing for orders to be delivered on-time, in-full and with the correct items.”

That is the number, and it comes from the company. It has not been audited, and Creekstone has not published the baseline or what the claims were costing. Even so, it is useful, because the shape of it is clear. Creekstone did not buy a tool to handle complaints faster. It fixed the step that was creating them.

This week’s brief is about that choice, and about the quieter work Baltz has been doing around it: rebuilding how product moves through the plant so there are fewer places for things to go wrong. It is worth your time even if you have never set foot in a meat plant. Every company that ships a physical product has a pack-off line somewhere, a moment where the work becomes the promise to a customer. Most of them have never looked hard at it.

THE OPERATOR

The situation

Creekstone started on a small farm in Campbellsburg, Kentucky, in 1995. In 2003 it bought the Arkansas City plant out of bankruptcy for $28.7 million. The plant had been built a few years earlier as one of the most modern beef facilities in the country, with cattle handling designed by Temple Grandin. A fire destroyed a major production area in 2013, and the company rebuilt with a $56 million expansion. In 2017 the Japanese trading company Marubeni bought Creekstone from a private equity owner. Marubeni has since kept investing, including a $35 million refrigerated distribution center with automated box storage. The plant can process about 1,400 head of cattle a day and employs more than 1,000 people. Kansas officials put Creekstone at about 1% of the American beef market. It is a private company, and the last sales figure it shared publicly was $651 million, in 2017.

It helps to understand what a beef plant sells. Customers don’t buy a cow. A steakhouse orders a certain cut, at a certain grade, within a certain weight range, trimmed a certain way, in a certain box, with a certain label. A grocer orders something different. An export customer in Asia orders something different again. The plant takes an animal apart and sorts the pieces into thousands of small promises, each with its own details.

Baltz calls the economics simple: “In the meat industry, specifically, yields are king.” Margins in beef packing are thin, and they depend on getting full value from every piece. A premium cut that ships at the wrong spec, or in the wrong box, or to the wrong customer loses more than its own price. The customer files a claim. Someone has to investigate it, issue a credit, and maybe replace the product. And the customer, who bought premium beef because they wanted a supplier they don’t have to double-check, starts double-checking.

The pack-off step is where much of this risk lives. It is fast, repetitive, cold work. People check what’s in front of them, choose the right box, and make sure the label matches the product. Baltz described what people across the plant check by hand today: specs, temperature, fat thickness, whether the meat is trimmed right. These are skilled people doing careful work. But a line that packs out 1,400 head of cattle a day will make some mistakes, and each one looks small until it reaches a customer.

That is the uncomfortable truth about many customer claims. They aren’t caused by carelessness. They are what you get when you ask people to be perfect at a task that repeats all day, every day.

The move

Baltz joined Creekstone in 2024, after working as a director of engineering at Tyson Foods. At Tyson, he had watched a camera system learn to tell one kind of chicken tray pack from another and send each to the right place. “That’s when I realized the power of AI,” he told MEAT+POULTRY this summer. “It becomes better than humans at identifying product.”

When he arrived, the plant already needed to update its pack-off system. The idea for how came from someone else. The plant’s director of engineering and maintenance at the time, Jaime Salinas, had heard about a small Nebraska automation company earlier in his career and brought it up. “So we went and checked it out,” Baltz said on a National Provisioner podcast in January, “and here we are.” The story starts with an engineer who knew the floor and knew what was possible, and a leader who listened to him.

What they built puts cameras and software at the pack-off line. The system sorts product automatically, checks each box against what should be in it, and records the details so any claim can be traced back to the box and the moment it was packed. Baltz described it in plain terms: the plant can now “do sortation, certain weights and primals for customers” and make sure “our product is on spec.” It works as a second set of eyes. People still run the line. The system catches the mismatch before it leaves the building.

The traceability part matters more than it might seem. Before, a claim was a question with no good answer: was the box wrong when it left, or did something happen on the way? Now the plant has a record of what went into every box. That changes a claim from an argument into a lookup. It also shows the plant which products, shifts, and lines make the most mistakes, so the next fix can go where it’s needed.

The team kept going. By March 2026, the plant was using the same approach to detect “leakers,” vacuum-sealed packages with air bubbles that would spoil the product or disappoint a customer. Leakers are another quality problem that is hard for a tired person to catch and easy for a camera to see.

The pack-off work is also part of a bigger rebuild. After his promotion to COO in July 2025, with a mandate to “drive productivity and automation across plant operations,” Baltz reorganized parts of the management team and redesigned how product moves through the plant. Creekstone replaced ground beef equipment that was more than 20 years old and rebuilt the room around it. It opened a new portion-cutting room in December 2025 that can cut to a quarter-ounce tolerance. The distribution center can now send up to 100 trucks out a day when needed. The ground beef line produces about 10 million pounds a year, and the plan is to reach 40 million within 12 to 18 months.

Baltz summed up the philosophy behind all of it in one line: “We spent a lot of time thinking intentionally about process flow. Fewer touches mean better quality and better economics.”

In that sentence, the AI plays a supporting role. The main job is designing a plant where product is handled fewer times, by fewer hands, with fewer chances to go wrong. The AI goes at the points where people still have to make fast, repeated judgments, and it helps them make those judgments right.

The result

The main result is the one Baltz reported: customer claims for order issues fell 75% after the pack-off system was fully running in September 2025. By his account, that means more orders arriving on time, complete, and with the right items.

The effects spread from there, even if Creekstone hasn’t put numbers on them. Fewer claims mean fewer credits and fewer replacement shipments. They also mean fewer hours spent looking into what went wrong, and fewer awkward calls with the customers who matter most to a premium brand. In a business where “yields are king,” a box that ships right the first time keeps its full value. The company also says the product-level data now helps it make better decisions and find more sales opportunities, though it hasn’t shared details.

The CEO, Yoshinobu Takahashi, framed the payoff in terms of customers, pointing to “the increased efficiencies” the system brings to Creekstone’s retail customers. That is the right way to look at it. The customer who stops getting the wrong box doesn’t care why it’s fixed. They just notice that ordering from Creekstone has gotten easier.

It helps to compare this with how companies usually handle complaints. The common approach works downstream: a better claims portal, a faster credit process, a chatbot to answer the angry email. Those tools make handling a mistake cheaper. Creekstone’s approach made the mistake itself less common. The first approach saves money on each complaint. The second removes complaints.

Now the honest take.

The 75% is a company figure, announced together with the technology partner, and nobody outside has checked it. Creekstone hasn’t said how many claims it had before, what they cost, or how the number is measured. A 75% drop from a small base is worth less than a 75% drop from a large one, and we can’t tell which this is. It is also one plant, measured over a few months, during a year when the company was changing many things at once: new management structure, new rooms, new equipment. The pack-off system is the most direct explanation for fewer order claims, but it may not be the only one.

The system also isn’t an overhaul of everything. It covers one step, bought from a specialist vendor and fitted to the plant. If you’re looking for a company that tore down its whole operation and rebuilt it around AI, this isn’t that. What Creekstone shows is something smaller and, I think, more useful for most leaders: how to pick the right step.

And we don’t know the cost. No investment figure has been published, so no one can calculate a payback. What we can say is that the result showed up in customer claims, which appear quickly and are hard to argue with, and not in a forecast.

The Craft of AI read

Most companies start their AI work by asking what the technology can do. Creekstone started by asking where its work goes wrong, and it answered with enough precision to point at one line in one room.

That order matters. A complaint is a signal that has traveled a long way from where it started. By the time it reaches customer service, it has been through a truck, a warehouse, and a customer’s receiving dock, and the cause is hard to see. The natural response is to put technology where the complaint arrives, because that’s where it shows up. Baltz put it where the complaint starts. I think that is the most transferable idea in this story, and it has nothing to do with beef.

The people in this story also deserve attention. The idea came from an engineer with years on plant floors. The system watches alongside people who already know what a correct box looks like, and it gives them a second set of eyes on the repetitive checks where anyone slips sometimes. Baltz is clear about how he sees it: the goal is “freeing up these hard and tedious tasks” so people can do work that adds more value. You don’t have to believe that line in every case to see that it fits this one. The line still runs on people’s judgment. The machine takes care of the part of the job that humans are worst at: doing the same thing perfectly for eight hours straight.

There is a link here to how Casella Waste made thirteen garbage routes disappear. Casella’s first visible win was something it could stop doing. Creekstone’s is similar: claims it no longer has to handle, because the mistakes behind them stopped happening. In both cases the most convincing evidence was work that went away. When a result removes work, you don’t need a forecast to believe it. You just need to count what’s no longer there.

The broader point is Baltz’s own: fewer touches. Every handoff, every time someone picks up the product, every relabel or re-sort is a chance for an error. Fix the flow first, so there are fewer of those moments. Then put intelligence at the few that are left. That order works in a meat plant, and it works in an office.

Things to consider

• Trace your complaints back to where they start, not where they land. Pull the last 100 customer claims or credits and ask, for each one, at which step the mistake was actually made. You will probably find that most of them trace back to a small number of steps. Those steps are where your first AI dollar belongs, well before the customer service desk.

• Look for the moment your work becomes a promise. At Creekstone it is the pack-off line, where a cut of beef turns into a labeled box for a named customer. Every business has one: the invoice going out, the order being confirmed, the shipment being released, the quote being sent. That moment usually gets less attention than the steps before it, even though any mistake made there goes straight to the customer.

• Give people a second set of eyes before you give them a replacement. Creekstone’s system checks and records. It doesn’t take over the line. That choice made the work easier to trust and easier to adopt, because the people on the floor still own the result. Before you ask what AI could do instead of your people, ask what it could double-check for them. The answer usually comes faster and costs less.

• Build the record, and claims become lookups. The quiet win here is traceability. Once every box has a record of what went into it and when, a customer claim stops being an argument and becomes something you can answer in minutes. It also shows you which product, shift, or line needs attention next. Ask how long it takes your team to answer “what actually happened with this order?” If the answer is hours or days, you are paying for that every week.

• Listen to the people who already know the floor. The idea that started all of this came from Creekstone’s director of engineering and maintenance, who had heard about the technology earlier in his career. Your best AI ideas may already be sitting with someone who has spent years watching the same process go wrong in the same ways. Ask them what they’d fix first, and take the answer seriously.

THE WORKBENCH

Do this tomorrow

One hour, a stack of recent complaints, and two people: one who handles the complaints and one who does the work the complaints are about.

Trace ten claims back to where they started. Pick ten recent customer complaints, credits, or returns, real ones from the last month. For each one, walk backward from the complaint to the step where the mistake was actually made: the order entry, the pick, the pack, the label, the invoice, the handoff. Write the step next to each claim. Don’t fix anything yet. Just mark where each one started.

Count the clusters. Most teams find that a few steps account for most of the ten. Circle the step that shows up most often. That is your pack-off line. Then ask the person who does that work a simple question: what do you have to check by eye, over and over, that you’d most like a second set of eyes on?

Put this year’s cost on it. For the circled step, add up what its mistakes cost you now: the credits issued, the replacement product and shipping, and the hours your team spends looking into and resolving claims. Use real numbers from this year’s records, not estimates about the future. Write the annual total on the board.

That figure is operating margin you are already losing, every week, at one step you can now name. It is also a much better starting point for your next technology conversation than a vendor’s list of features.

THE QUESTION

Andrew Baltz didn’t start with a plan to reinvent a beef plant. He started with a pack-off system that needed updating, an engineer who knew a better option, and a clear view of where mistakes were being made. A few months after the system was fully running, order claims were down 75%, by the company’s count.

So here is the question to take into your week. If you traced your last hundred customer complaints back to where they began, how many different places would you find?

If the answer is dozens, your problem is spread across the whole business, and no single tool will fix it. You will need to look at how the work flows and cut down the number of touches, the way Baltz did with the ground beef room and the portion room. If the answer is a handful, you are in a better spot than you think. You have found your pack-off line, and you can start there.

Either way, the place to start is the step where the mistake happens, not the desk where the complaint arrives.

---

Want this done for you?

You just traced ten complaints back to the step where they start, and put a price on it.

My Ground-Up Workshop is the rigorous version of the same exercise, run on your business with the small group of your people closest to that workflow. Every AI strategy you’ve been sold starts with software. This one starts with the people who do the work.

One to two days in person, then a synthesis week, and you walk out with a plan: a ground-truth map of how your core workflow actually runs, the one where a rebuild would release real operating margin, a target operating model with AI built in from the ground up, and a 30-60-90 day plan your own team can start on. I do a small number of these a quarter, for $20,000.

I’m a Partner at geniant. Since 2003, our craftspeople have done one thing: understand how work actually happens, design how it should happen, and build software that works the way people do. When a workshop gives you the roadmap to execute or turns into a build, they’re who does it: one senior-led team, layered on top of the systems you already run, value in weeks rather than months.

Book a discovery call →

— Grant K. Baldwin