The Death of Top-Down Planning: Why Aggregation is a One-Way Street

The Uncomfortable Truth

For 30 years, we've been solving the wrong problem. We've spent billions on forecasting engines, demand sensing algorithms, and machine learning models—all designed to make our aggregate forecasts more accurate. And we've succeeded. Portfolio-level forecasts have never been better. But here's the uncomfortable truth: When it's time to execute, the mathematics abandon us.

You cannot manufacture a "Product Family." You cannot ship a "Regional Aggregate." You cannot load a "Brand Category" onto a production line. The moment you need to convert that beautiful aggregate forecast into which SKU, on which machine, on which day—the Central Limit Theorem that made your forecast look so stable provides zero help.

This is the Disaggregation Wall, and it's where every Top-Down planning system in the world quietly breaks down. The industry's response? Heuristics. Historical proportions. Planner judgment. Sophisticated guesswork dressed up as methodology. What if there's a better way? What if we stopped trying to reverse a mathematical operation that cannot be reversed—and instead built the plan from the ground up? Let's challenge 30 years of conventional wisdom—because the mathematics demand it.

Why Aggregates Look So Good

For decades, Supply Chain Planning has relied on a comfortable lie: Plan at the top, execute at the bottom. Forecast the product family. Forecast the region. Forecast the quarter. The numbers look clean. The executives are happy. The variance is low. Then reality hits: You need to tell the factory which SKU to make, on which line, on which day.

The Statistical Magic Behind Aggregation

When you forecast at the aggregate level—across product families, regions, or time periods—you benefit from two fundamental statistical principles:

The Law of Large Numbers

When you combine many independent forecasts, extreme errors tend to balance out.

Over-forecast SKU A by 20%?
Under-forecast SKU B by 18%?
Portfolio-level error: Much smaller

For non-mathematicians: If you flip one coin, predicting heads or tails is a guess. But if you flip 1,000 coins, you can confidently predict close to 500 heads. The aggregate is stable even when each individual outcome is random.

The Central Limit Theorem (CLT)

More precisely: When you sum demand across many SKUs (each with its own forecast error), the distribution of total demand becomes increasingly normal and predictable—even if individual SKU demand is erratic. The aggregate forecast error, as a percentage, shrinks as you add more SKUs.

The result: Executives love portfolio-level forecasts because they work.

"We'll sell 1 million units of Product Family X next quarter"
"European demand will be €50M"
"Brand Y will grow 8% YoY"

At this level, the forecast looks beautiful. Confidence intervals are tight. Variance is low.
The trap: You cannot execute at this level.

The Disaggregation Wall: Where Mathematics Meets Reality

In the physical world, you cannot:

Manufacture a "Product Family"
Ship a "Brand Category"
Load a "Regional Aggregate" onto a production line you must execute at the atomic level:

Specific SKUs
Specific machines
Specific warehouses
Specific days or shifts

This is where Top-Down planning hits the Disaggregation Wall—the moment you must convert your statistically-sound aggregate forecast into executable instructions.

The Unsolvable Mathematical Problem

The question every planner faces:
How do you split "1 million units of Product Family X" into:

SKU A: 247,382 units?
SKU B: 318,926 units?
SKU C: 433,692 units?

There is no mathematical theorem that tells you how to reverse the aggregation. The Central Limit Theorem explains why the sum is predictable. But it provides zero guidance on how to decompose that sum back into its components when market conditions are changing.

The Heuristic Trap

Without mathematics to guide you, the industry resorts to heuristics:

Common Disaggregation Methods:

1. Historical Proportions

"SKU A was 24.7% of the family last year, so allocate 24.7% of next quarter's forecast"

2. Equal Distribution

"Split the volume evenly across all active SKUs"

3. Planner Judgment

"I think SKU C will be hot this quarter based on the sales meeting"

Why This Inevitably Fails

These heuristics assume tomorrow looks exactly like yesterday. They embed systematic bias because:

• Product mix shifts over time (new SKUs launch, old ones decline)
• Promotions change demand patterns unpredictably
• New SKUs have no historical proportions to reference
• Seasonality varies by SKU, not just by product family
• Customer preferences evolve continuously

The moment market reality diverges from historical patterns—which it always does—your disaggregation logic breaks.

The Downstream Consequences

You're left with:

• Excess inventory of the wrong SKUs
• Stockouts of the SKUs customers actually want
• Manual firefighting to rebalance the plan
• Finger-pointing between Demand Planning and Supply Planning
• Millions in working capital tied up in the wrong places

The Mathematical Irony

The same statistical principles (Law of Large Numbers and Central Limit Theorem) that made your aggregate forecast look stable are completely useless when you need to execute.

The error-canceling magic works in one direction only:

• Aggregation up (from SKU to Family): Mathematics guarantees stability
• Disaggregation down (from Family to SKU): No mathematical principle exists

Aggregation is a one-way street. This is not a limitation of current technology or methodology—it's a fundamental mathematical reality.

The Alternative: Bottom-Up Planning

What if, instead of planning at the top and disaggregating down, we planned at the bottom and aggregated up?

The Bottom-Up Paradigm

• Plan every SKU × Location × Day
• Let executives view aggregates by summing the vectors
• Aggregation becomes a display layer, not a planning methodology

The Benefits

• 1. No Disaggregation Error: Volume was built from the ground up—no guesswork required.
• 2. No Systematic Bias: Each SKU is planned based on its own demand signal, not forced into family-level proportions.
• 3. Mathematical Integrity: The sum of the parts always equals the whole—by construction, not by reconciliation.
• 4. CLT Still Works: Executives still benefit from error-canceling when viewing aggregates—but now the underlying atomic plan is rigorous.

The Challenge: You're now planning millions of data points instead of thousands. This is not a trivial computational problem—but it's a solvable one.

When We Do "Cheat" (And Why It Sometimes Works)

We must be honest: Sometimes, when SKU-level data is too sparse, we do use top-down disaggregation as an enrichment strategy. Think of it as playing an informed game of roulette.

When Top-Down Can Help

For new SKUs with no demand history, or for long-tail products with intermittent sales, bottom-up forecasting can produce nothing useful. In these cases:

• We start with a top-down allocation based on historical product mix
• We treat this as a prior (in Bayesian terms), not as truth
• As actual demand signals emerge, we immediately replace the heuristic with real data

Sometimes this actually yields better short-term results than leaving the plan blank or relying purely on sparse signals.

The Critical Caveat

But it's essential to understand: This is not guaranteed to work. Why? Because it assumes the product mix will never change—which is obviously false.

• If SKU C is growing faster than historical proportions suggest → under-allocated
• If SKU A is in decline → over-allocated
• If a new trend emerges (sustainability, flavor preferences, regional shifts) → historical ratios are useless

The fallacy is easy to see: If product mix were truly stable, you wouldn't need planning—you'd just repeat last year's production schedule. Our Philosophy: We use top-down disaggregation as temporary scaffolding, not as a foundation. The moment we have enough SKU-level signal to plan bottom-up, we discard the heuristic. The goal: Minimize our dependence on assumptions that we know will eventually fail.

The Path Forward

The mathematical argument is clear: Aggregation is a one-way street. Top-down planning works beautifully for portfolio-level visibility. But when execution demands atomic-level precision, the mathematics provide no path backward.

Bottom-up planning is the only rigorous solution. If you're an academic, solution architect, executive, or planner wrestling with the disaggregation wall in your organization, we'd welcome a detailed technical discussion.