Cost per case won CPG the last decade. Cost per decision wins the next.

The co-packer's choice that should never be made

A CPG founder preparing to launch learned what happens when manufacturers prioritize instinct over evidence. A co-packer had committed to capacity: six SKUs across two unique formats, at minimum order quantities a startup could sustain. Launch dollars were committed. Media was bought. Then the manufacturer called: the capacity was being reallocated to a legacy brand, deemed safer and more predictable amid chaos.

The question was simple: what was this decision based on? What calculation could prove otherwise? The honest answer was that none existed in a form anyone could produce. Safer and more predictable were judgments formed from experience, not calculations anyone could defend on the merits.

The founder won three SKUs back. Not because the forecast was stronger, the margin math was better, or the facts favored them. The founder won because they were in the room and they were convincing.

That was not a COVID problem. It was a decision problem.

Every week, across co-packers, private label manufacturers, and internal plant networks, constrained capacity goes to the most persuasive pitch rather than the most defensible one. The decision latency is real. It leaves manufacturing and commercial operating in separate information systems, with no shared understanding of what the business can actually promise.

The same disease, a hundred times the scale

A customer asked one of the largest beverage manufacturers in North America a straightforward question: can you take more volume?

The answer should have taken an afternoon. It took nine days.

Manufacturing knew what the lines could run. Supply chain knew what the network could move. Finance knew what it would cost to serve. Commercial knew what the customer would accept. Four correct answers. Four different systems. Four slightly different definitions of the words in the question. Nearly two weeks spent reconciling them into something a leader could sign.

They reached a good answer, too late to negotiate from strength.

The term for what this costs is organizational latency: the delay between a business knowing something and being able to act on it. Every other latency in CPG is measured to the second. This one is not measured at all.

The evidence that organizational latency is not a niche complaint comes from leadership surveys. In McKinsey’s global survey on decision making (fielded 2018, 1,259 respondents), only about one-fifth of executives said their organizations were good at making decisions. Fewer than half said decisions got made quickly. And barely a third said their decisions were both high quality and fast. Separately, McKinsey found that nearly three-quarters of senior executives thought poor strategic decisions were at least as common in their companies as good ones.

Cost per case is necessary. It is no longer sufficient.

Thirty years of relentless discipline drove cost per case toward its theoretical floor. Cost per liter, per pallet, per line hour, per delivered mile all measured to the fourth decimal. That discipline still matters. In high-volume commodity categories, it will still decide who survives a bad year.

What changed is that it no longer separates anyone. A competitor can buy the same high-speed line, the same lightweight preform, the same automation stack and a comparable freight contract inside eighteen months. The engineering is no longer proprietary. It is a procurement exercise.

The evidence is on the shelf. Circana reported US private label sales at $330 billion in 2025, around a quarter of units sold. Retailers are matching quality at lower price points, in the same plants, often with the same co-packers. McKinsey’s outlook on CPG growth describes an industry entering 2025 on the back of weak volumes and continued margin pressure, with consumers still hunting value and increasingly trusting store brands. Cost leadership has become the price of entry rather than the source of advantage.

Meanwhile, the decisions that actually move earnings have migrated out of the plant and into the space between the plant, the lane, and the customer conversation. Which brand gets the line. Who receives constrained supply. Whether this promotion’s incremental cases are worth their margin. Whether a quality drift is an operational nuisance or a commercial liability.

None of those are engineering problems. All of them decay with time. A correct answer that arrives after the customer has committed elsewhere is worth precisely nothing.

From cost per case to cost per decision

Organizational latency is the disease. Cost per decision is the diagnostic.

Not the financial impact of a decision, the organizational cost of reaching it.

How many analyst hours. How many meetings convened to reconcile reports that disagree. How many versions of the spreadsheet. How many days between a signal appearing in the data and someone acting on it. And then the part almost nobody counts: how often the decision proved right, and how often it landed in time to be worth making.

That last piece is what stops this becoming another efficiency metric. A cheap decision is not one that consumed fewer analysts. It is one that reached the right answer before the window closed. Speed alone produces fast, confident and wrong, which is the trap most AI programs in this industry are currently walking into.

Cost per case is a stock: improve it and it stays improved. Cost per decision is a flow. Every day the organization does not improve it, the company pays again.

The factory is your largest commercial sensor, and most companies are still reading it as a cost report

For thirty years, manufacturing’s mandate was to take cost out. It worked. But a modern plant now generates more signal about the near-term commercial future of the business than any other part of it.

Line speed and yield tell what the business can actually promise. Changeover patterns tell what the promotional calendar is really costing. Reject rates tell which customer relationships are about to come under strain. Downtime tells which service commitments are at risk this week, not next quarter.

Every one of those is a commercial input. Almost none reach a commercial conversation in time to change it. The factory has quietly become the company’s largest sensor array for growth decisions. Most companies are still reading it as a cost report.

The instinct right now is to deploy AI inside each function. An assistant for manufacturing, one for finance, one for procurement, one for commercial. Each is individually defensible. Each has a sponsor and a business case. Collectively, they raise cost per decision.

They raise it because each reimplements its own definition of OEE, its own customer hierarchy, its own fiscal calendar, its own idea of what a case is. Four functions do not produce four accelerated functions. They produce four confident systems that disagree with each other, sitting on top of the reports that already disagreed.

MIT’s Project NANDA reported that around 95% of enterprise generative AI pilots showed no measurable effect on the P&L, and attributed the failures not to model quality but to brittle workflows, misalignment with daily operations, and tools that could not retain context. Separately, Gartner forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Not model failure. Management failure.

What is missing is organizational memory

Most of the intelligence a company needs is already inside it. Almost none of it is connected.

McKinsey’s research on scaling agentic AI puts a number on the gap: nearly two-thirds of enterprises have experimented with agents, but fewer than one in ten have scaled them to deliver tangible value. Eight in ten name data limitations as the roadblock. The specific failure will be familiar to anyone who sat through a reconciliation meeting: different systems defining the same record differently.

Organizational memory is the missing piece. One shared understanding of what the business means by its own words, its metrics, its customers, its products, its rules, that every function reasons against. Not another assistant. The thing underneath all of them.

What changes when it exists: a production issue becomes a customer conversation the same morning, not the following week. A supplier’s slipping performance becomes a service prediction with account names attached. A proposed promotion becomes a manufacturing implication before anyone commits to it, rather than after. An allocation call between two brands becomes a comparison anyone can see, rather than a judgment the losing party can only respond to with persuasion.

The economics are the real argument. The first capability built this way costs more than building it in a silo, noticeably more, which is why it is hard to fund. The second costs a fraction. The fifth costs very little. The return comes from reuse across a portfolio of decisions, never from optimizing one.

Where it shows up in the P&L

Fair challenge from any CFO: does this move earnings, or is it an operating improvement that never reaches the statement?

Every point of margin begins as a decision. It reaches the statement through three terms, each with a different owner.

The three margin levers

Every term has a named owner, and every one is currently gated by how long the organization takes to decide anything. That framing is worth more to a finance team than a benchmark borrowed from someone else’s plant network.

McKinsey’s decision-making research also points to a return worth noting: organizations that made high-quality decisions quickly and executed them quickly reported better growth and returns than their peers. Decision speed is not a soft benefit. It shows up.

Start with one decision

Not a platform. One decision.

Pick something recurring the organization does slowly or badly: a production reallocation, a promo depth call, a service commitment to one customer group. Measure what it costs today: elapsed days, hours consumed across every function that touches it, how many past instances proved right and got acted on.

Most companies have never measured this. The number is usually worse than leadership expects, and that discomfort is the most useful asset in the program.

Then build the thinnest slice of organizational memory that improves that one decision, with a named business owner and a result finance will certify. Fund the next phase on that number.

The future of competitive advantage

In the next few quarters, the largest account will ask whether the company can support another thirty million cases. It is the same question that took nine days.

The factory is already fast. The organization is not. That is the bottleneck.

The version of that conversation worth building toward is one where the organization answers while they are still in the room, because manufacturing, supply chain, finance, and commercial are reasoning from the same understanding of what the business can actually promise. At that point the company has stopped selling capacity and started selling certainty. Capacity is a commodity with a market price. Certainty is not, and no competitor buys it with an eighteen-month capital plan.

In the era of cost per case, companies won through operational excellence. In the era of cost per decision, they win through organizational clarity. The companies that win the next decade will not replace judgment with AI, judgment is the part worth keeping. What they will replace is persuasion with evidence that arrives fast enough to still be worth arguing about.

So it is worth asking the leadership team this quarter: what does a decision cost us today, and whose name is on that number?