EXECUTIVE SUMMARY

Two traps, one blind spot

Supply chain leaders today are caught between two flawed instincts. The first is a comfortable but outdated habit: classifying inventory by value alone, the way it has been done for decades. The second is a fashionable overcorrection: the belief that more data and cleverer algorithms can forecast away uncertainty entirely. Both miss the same thing — how demand actually behaves.

The sources behind this report make the case for a third path. Rather than ranking products only by revenue contribution, or outsourcing judgment to an AI forecasting engine, companies should characterize demand by its underlying pattern — using the Average Demand Interval (ADI) and the squared Coefficient of Variation (CV²) — and compress lead times, which do far more to tame supply chain volatility than incremental forecasting gains ever will.

*Illustrative, based on the bullwhip dynamics discussed in this report.

What this report covers

—   The relics of strategy —  why classic ABC analysis, for all its simplicity, systematically misclassifies risk.

—   The AI trap —  why sophisticated forecasting models can be a category error in a supply-constrained world, and why lead time — not forecast precision — is the real lever.

—   A smarter path —  how ADI and CV² reveal four distinct demand patterns, each requiring its own playbook.

—   From prediction to navigation —  a practical shift in mindset, from point-forecasting to scenario-based, constraint-aware planning.

SECTION 01

The relics of strategy: the problem with ABC analysis

Traditional ABC analysis, rooted in the 80/20 Pareto principle, remains one of the most frequently used classification techniques in industry today. It segments products primarily by value or sales volume, often assigning 20% of items to category A, 30% to category B, and 50% to category C.

It is easy to see why the method has survived so long: it is simple, intuitive, and requires nothing more than a sorted spreadsheet. But that simplicity is also its undoing.

FIGURE 1.  Classic ABC segmentation ranks SKUs purely by usage value — the top ~20% of items typically account for roughly 80% of total value.

The flaw hiding in plain sight

ABC analysis focuses solely on financial impact while ignoring demand volatility. A high-value "A" item with stable, predictable demand — call it AX — requires a completely different replenishment strategy than a high-value item with erratic, "lumpy" demand — an AZ item. Treat them the same, using the same reorder logic and the same safety-stock formula, and the result is entirely predictable: bloated stock sitting alongside empty shelves.

This is a hallmark of what is known as the bullwhip effect, where distorted information moving up the supply chain leads to massive amplification of variance at each successive tier.

FIGURE 2.  Overlaying demand variability (X/Y/Z) onto value tiers (A/B/C) exposes the AZ cell — high value, erratic demand — as the segment classic ABC analysis handles worst.

SECTION 02

The AI trap: precision vs. physics

There is currently a heavy industry focus on using AI for forecasting, under the belief that better mathematical techniques can "solve" product uncertainty. The sources argue this is often a category error: a precision problem is being thrown at what is, underneath, a physical and structural constraint problem.

Regime shifts break the fundamental assumption

AI and machine learning models generally rely on the assumption that historical demand is predictive of future demand. But in a "supply-driven" market — one where capacity limits and geopolitical blocs set the boundaries of what is possible — past patterns become far less relevant. No amount of model sophistication compensates for training data that no longer describes the world.

Lead time, not forecast error, is the dominant lever

A major cause of supply chain deficiency is not just poor forecasting — it is long lead times. Companies routinely invest heavily in AI to shave a few percentage points off forecast error, while ignoring lead-time compression, which has a far more dramatic effect on reducing the bullwhip effect. Increasing lead times magnify variability, yet many firms continue to treat lead time as "noise around a mean" rather than the structural constraint it actually is.

FIGURE 3.  The bullwhip effect: small, ordinary shifts in real consumer demand become large, costly order swings by the time they reach manufacturers and suppliers.

FIGURE 4.  Illustrative comparison of two common improvement levers — cutting lead time typically dampens variance amplification far more than incremental forecast accuracy gains.

MRP's quiet failure mode

Reliance on Material Requirements Planning (MRP) often fails for a related reason: it uses outdated values and assumes a smooth, linear supply response that no longer exists in a constrained infrastructure market. The system isn't broken — the assumptions feeding it are.

SECTION 03

A smarter path: characterizing demand with ADI and CV²

Instead of asking only how much value a product brings — the ABC question — smarter practice asks how the product is actually consumed. Two metrics do this well: the Average Demand Interval (ADI) and the Coefficient of Variation squared (CV²).

—   ADI (Average Demand Interval) —  measures the average interval between two non-zero demand occurrences. A high ADI means demand shows up rarely.

—   CV² (Coefficient of Variation squared) —  measures the volatility, or "lumpiness," of demand sizes when it does occur.

Plotting products against these two axes produces four distinct demand patterns, each calling for a different operational response.

FIGURE 5.  The ADI / CV² framework sorts demand into Smooth, Erratic, Intermittent, and Lumpy patterns — a classification based on behavior, not value.

Matching strategy to pattern, not forcing one model on everything

Rather than relying on a one-size-fits-all AI forecast, companies can use these patterns to determine what should be produced, and when. "Smooth" (AX-type) items can be managed via lean, automated JIT systems. "Lumpy" or "Erratic" items, by contrast, require dynamic safety-stock buffers rather than pure mathematical prediction — because for that segment of the portfolio, precision forecasting is chasing a signal that, structurally, isn't there to find.

CONCLUSION

From prediction to navigation

The sources are direct on this point: smooth is smart. To tame the bullwhip effect and reduce cost, companies need to stop trying to forecast their way out of what are, fundamentally, physical constraints.

That means three concrete shifts in how planning teams operate:

—   1. Move from point-forecasts to scenario ranges —  anchored to real, current constraints rather than historical patterns that a supply-driven market has already made obsolete.

—   2. Adjust replenishment orders fractionally —  to avoid over-reacting to short-term noise, which is one of the primary mechanisms feeding the bullwhip effect.

—   3. Prioritize lead-time reduction over mathematical precision —  since compressing the pipeline typically does more to stabilize the whole system than any incremental forecasting improvement.

A note on how to use this framework

None of this argues for abandoning analytics — it argues for pointing them at the right question. ABC still has a role in prioritizing management attention by value. AI forecasting still has a role where demand truly is smooth and history truly is predictive. The shift this report recommends is sequencing: characterize demand behavior first, using ADI and CV², before deciding which tool — JIT, dynamic buffers, or a forecasting model — actually fits the pattern in front of you.