Growing Fast Without the Inventory Tax: How Carbliss Rebuilt Its Forecast

In 2024, Carbliss went looking for a 100,000-square-foot warehouse. The Wisconsin cocktail brand had run short of its best sellers the year before, raised capital, and built enough inventory to make sure it never happened again. It over forecasted, and now it needed somewhere to put the cans.

Supply Chain Now host Scott Luton opened “Playing Offense: Growing Fast Without the Inventory Tax” with the question Carbliss had lived through: “What happens when your business starts growing faster than your ability to forecast demand?” When growth comes from new flavors and new states, a forecast built from last year’s sales has very little to work with. Carbliss fixed it by giving its forecast the context its own people already carried, and by proving the new numbers against the old process before asking the team to rely on them.

This article recaps the webinar, in which Casey Liddicoat, Carbliss’s VP of Operations and Quality, and Omnifold CMO Garrett Long joined Luton and co-host Scott DeGroot.

What does 100x growth do to a supply chain?

It multiplies the number of planning decisions far faster than a team can grow to handle them. Carbliss, a premium ready-to-drink canned cocktail brand founded in 2019 by a husband-and-wife team in Plymouth, Wisconsin, went from roughly 40,000 cases to more than 4 million in four years, and from six flavors to more than 32.

“There’s no playbook for 100x-ing your output in four short years,” Liddicoat said.

The brand grew deliberately, expanding backyard to backyard in a model Liddicoat credits to Carbliss’s founder Adam Kroener. “Our mantra around here is an inch wide and a mile deep,” he said. Even so, each new region brought its own flavor preferences, and “with that just comes tremendous complexity.”

DeGroot, who spent more than 30 years in supply chain leadership at Kimberly-Clark, explained why that complexity compounds. Every added item, version, or size multiplies the combinations a planner has to cover, and “the planning problem becomes exponentially complex.” Long put a number on it for Carbliss: “You take 32 SKUs times six pack sizes times all the places you want it to be, and all of a sudden you’ve got a billion data points to analyze.”

New products make it harder still. Once a brand saturates a market, Long said, new revenue comes from new products, and “that growth by definition is coming from something you don’t have history for.” DeGroot agreed from experience: “New product launches are always, in my opinion, the most disruptive.” Liddicoat compared a launch to betting at the racetrack. “You think you might have it all figured out and everything can point in one direction, but there always inevitably seems to be something in that race that changes your opinion halfway through.”

Why did the spreadsheet forecast stop working?

Carbliss’s forecast was a negotiated number, set with commercial. Liddicoat described the process plainly: “We got a bunch of folks in a room. We looked at the data. The sales team primarily slapped a growth figure on that they thought was attainable.” The spreadsheet went around by email a few times until the group settled on a figure. That number fed the business plan and budget, and the company signed production contracts with its co-packers to match.

“It wasn’t very scientific,” he said. “We never really refreshed that forecast either.” He put the cost to his team as a question: “Would you go and pick a stock tomorrow if I only handed you the newspaper from nine months ago?”

The plan also couldn’t change course, a serious handicap in beverage alcohol. Luton noted that the industry faces regulatory hurdles “much more complicated than many, many other sectors.” A launch can slip when a state registration isn’t approved in time or a chain program underperforms. “That static model really did not allow us to be nimble or pivot whatsoever,” Liddicoat said.

DeGroot pointed to a second weakness in a consensus number. “Human beings are wonderful, but they are also biased,” he said. Sales leans toward its revenue targets, supply chain leans away from excess inventory, and a figure negotiated between them inherits both pulls. Carbliss was far from alone in planning this way: Long noted that companies with revenue in the billions still run on spreadsheets.

What did the inventory roller coaster cost Carbliss?

It cost Carbliss sales in one year and cash and warehouse space the next. In 2023, like most young brands, the company was limited by “access to cash,” as Liddicoat put it. “We picked the SKUs that we thought we were going to need the most of, and we really bet on them. We got some right, we got some wrong.” More of them ran out than sat too long.

New products were hit hardest. During peak season, some new flavors were out of stock for more than 30 days, forcing emergency purchase orders and supply diverted from other SKUs. Wholesale distributors were “really upset by the fact that they’ve got something that’s making them a lot of money, and then we just cut it off,” Liddicoat said. “It’s very stressful, very strenuous on the team.”

So the company secured more capital and built “a mountain of inventory going into 2024.” In Liddicoat’s words, “We were a little too bullish.” The overcorrection brought aging stock, expensive capital tied up on the books, and the search for that warehouse. The plan was also national, with little regional detail, and “there were a lot of freight inefficiencies as a result.”

DeGroot said the pressure peaks at exactly this stage of growth, when a company is “searching for working capital to make payroll, to fund inventory growth.” Coming into 2025, Carbliss reached its turning point. “If we don’t figure this out, this is going to continue to be a significant detriment or really hamper our growth,” Liddicoat said. “This is a very, very costly problem and it has to be solved.”

How do you forecast products that have no sales history?

You give the forecast what the business knows beyond its sales history. Long calls the missing information the Context Gap, and he describes it in three parts.

The first is history, the record of what has sold. Every company should use it, but a new flavor or a new state has none. The second is structure, meaning how the business works. Any planner knows without thinking that a canned cocktail behaves differently from a T-shirt. “Traditional planning systems have none of that,” Long said. “It’s just a formula that says, give me data and I’m going to project.” The third is knowledge, meaning what people in the business know is changing: a sales rep hears an account plans to order more, or a distributor asks for less.

Context

What it covers

At Carbliss

Where traditional planning stops

History

Past sales, orders, and inventory

Shipments by SKU and distributor

A new flavor or state has nothing to extrapolate

Structure

How the business works: products, network, constraints

Flavor profiles, pack sizes, co-manufacturers, one-day transit to customers

Rarely represented; the method projects history forward

Knowledge

What people know is changing

Account conversations, distributor requests, state registration timing

Typed in as a manual override, or left out

Long’s second point concerned the math itself. “For the last 20 years, we’ve been trained to just accept one of the six dropdowns from our planning system,” he said. “We don’t need to accept that anymore because the details matter.” Carbliss’s supply chain behaves differently even from another cocktail brand’s, so its forecast should come from a model developed for its own business.

For Carbliss, Omnifold trained that model on the brand’s sales history, its co-manufacturing network, and demographic data by zip code. According to the Carbliss case study, the model reached 80% accuracy at SKU and distributor level in its first month in a state with no sales history. Across the pilot months, accuracy at the same level ran between 72% and 74%, against 28% to 37% for the brand’s prior method.

How did Carbliss’s planners come to trust the new numbers?

They watched the new forecast outperform the old process on Carbliss’s own data for several months before using it day to day. Liddicoat treated trust as the central obstacle. “When we ask somebody to do something different than they’re accustomed to, we’re asking them to change their identity a little bit,” he said. “We derive our value, our personal value, from the things that we do and how we contribute.”

So the team tested before switching. Using historical data, Carbliss rewound the clock, measured how accurate its existing process had been, and set Omnifold’s results beside it. “Over a course of several months, the team could become comfortable with those results,” Liddicoat said. “Quite frankly, they wanted to be a part of it.”

Long said the test was tougher than Liddicoat let on. “Casey’s underselling how rigorous he wanted to be, and it’s really fair,” he said. “We have no place to say just trust us, because ultimately we’re asking teams to change the center of what they do.”

DeGroot credited the time Liddicoat invested in that process. “Experiential learning builds trust,” he said. “The math can be super correct, but if people don’t use it or don’t trust it, it might as well not be running.”

What changed when the team started talking to the AI like a colleague?

Use of the platform took off once the team stopped issuing step-by-step commands and started giving it context. After ten or more attempts at one problem using commands, Liddicoat suggested another approach. “What if this thing is smarter than we think it is and we just talk to it like a person and just give it the context rather than tell it how to achieve an output?” he said. “That’s when the light bulb moment went off, and it was like, ‘Oh my gosh, this is just another person, another part of our team.’”

Long compared it to having an operations researcher on staff, and described watching a planner on Liddicoat’s team work through an expansion into a new state with the platform in a short amount of time. Planning, he said, is “a really hard and mostly thankless job,” and supply chain leaders know their planners rarely enjoy it. “We think we can help the planner smile,” he said.

Adoption spread past the planning team. Sales analysts started asking the agent to structure data and build charts for them, which “wasn’t part of the original scope,” Liddicoat said.

Liddicoat has seen plenty of AI claims. At the American Manufacturing Summit in Chicago, he said, “eight out of 10 booths were AI based this or that.” Luton raised the concern many buyers share, “AI washing,” meaning products sold as AI that aren’t, and asked Liddicoat directly whether Carbliss’s experience was the real thing. “It is!” Liddicoat exclaimed.

What results has Carbliss seen?

Carbliss has cut its inventory sharply and has had no stockouts caused by planning in the past year. Liddicoat said the company once carried as much as 300 days of inventory in some cases. It set a target of 45 days on hand, “and I’m happy to report we’re pretty darn close to that,” he said, holding there “consistently throughout the year” despite seasonal beverage demand.

“We’ve had zero stockouts in the past year that have been a result of planning,” Liddicoat said. He was careful about the limits of that claim: execution problems still happen in a network that runs through several co-packers, but the team now responds to them faster.

The freed capital went back into growth. “It’s allowed us to press the gas pedal down a little harder in expanding into more states,” he said. Less inventory also meant lower storage charges. Carbliss now plans by region, aiming to keep stock within 500 miles of each destination for one-day delivery, and the shorter freight miles help offset rising lane rates. “We’re much more efficient with our capital, and we’ve seen some expense reduction as well,” Liddicoat said.

DeGroot read the results the way a CFO would. Product in stock protects revenue. Transportation is “the most variable cost,” so fewer freight miles go straight to margin. “My CFO likes that,” he said. With inventory near 45 days on top of that, “those are difficult things to do, and you’re doing all three.”

Next, as distributors gain confidence in Carbliss’s accuracy, Liddicoat expects the company to “earn the right to suggest their ordering.” For Long, results like these are why Omnifold applies AI to optimizing the supply chain. The goal, he said, is “putting capital back into the business.”

What should other fast-growing companies take from Carbliss?

Long’s advice was to stop accepting a planning approach that only roughly fits the business. Supply chain leaders, he said, “can point to cash they wish they had,” even after investing tens of millions of dollars in planning technology. No company would let its packaging team pick from six stock boxes that don’t fit the product, and planning no longer has to work that way either. He also recommended asking for proof on your own data before committing, as Carbliss did.

DeGroot called planning the company’s “central nervous system,” since every purchase and production run starts with its signal. If a better version is available and a company doesn’t use it, he said, “we’re really doing a disservice to the people who are buying things and the people in our factories making the things.”

Carbliss’s own test is a reasonable place to start. Set last year’s forecast beside what actually shipped, find the biggest misses, and check how many of them involved a new product, a new market, or something your team knew that never reached the forecast.

Frequently asked questions

How did Carbliss forecast demand before Omnifold?

Carbliss built an annual forecast in a spreadsheet. A cross-functional group reviewed the data, the sales team applied a growth figure it considered attainable, and the number became the basis for the budget and co-packer production contracts. It was rarely refreshed during the year.

How do you forecast demand for a new product with no sales history?

Use the information the business has beyond sales history: the structure of the business, such as products, pack sizes, and distribution network, and the knowledge people in the business hold, such as account plans and distributor requests. Carbliss’s model reached 80% accuracy at SKU and distributor level in its first month in a new state.

What is the Context Gap?

The Context Gap is Omnifold’s term for the information a forecast needs but traditional planning leaves out. It has three inputs: history (what has sold), structure (how the business works), and knowledge (what people know is changing).

How much did Carbliss reduce inventory?

According to Casey Liddicoat, Carbliss went from carrying as much as 300 days of inventory in some cases to running close to its 45-day target throughout the year. Over the same period, it had zero stockouts caused by planning.

How did Carbliss’s team come to trust the new forecast?

Carbliss backtested for several months. The team compared how accurate its existing process had been on historical data with the results from Omnifold, and adopted the new forecast once the team was comfortable with what it saw.

Whitepaper cover: AI research for supply chain planning

FREE WHITEPAPER

The next leap in supply chain AI isn’t automation.

It’s better decisions.

Omnifold’s researchers, from OpenAI, Google, MIT and Stanford, show how reasoning AI makes forecasts more accurate, and the impact to inventory, expedites and margin.

—

Why constant disruption demands more than historical data

—

How reasoning AI forecasts new products with no sales history

—

The seven capabilities behind better planning decisions