What Is XYZ Analysis for FBA Inventory?
XYZ analysis classifies your SKUs by how predictable their demand is, not how much they sell. While ABC analysis answers "which products generate the most revenue," XYZ analysis answers "which products can I actually forecast reliably?"
The classification uses the coefficient of variation (CV) of demand over time:
- X items: Stable, predictable demand (CV < 0.5). These SKUs sell roughly the same quantity week to week. Standard statistical forecasting methods work well here.
- Y items: Variable demand (CV 0.5 to 1.0). Sales fluctuate noticeably, often driven by promotions, seasonality, or competitor activity. Forecasting works but requires wider safety stock buffers.
- Z items: Erratic, unpredictable demand (CV > 1.0). Sales are sporadic or wildly inconsistent. Traditional forecasting fails here; you need specialized methods like Croston's.
The real power of XYZ analysis comes from combining it with ABC tiers into a 9-cell matrix. An AX item (high revenue, predictable) gets a completely different reorder policy than an AZ item (high revenue, erratic) or a CZ item (low revenue, erratic). This matrix is the foundation of differentiated inventory management.
XYZ Analysis Formula: Coefficient of Variation
The coefficient of variation measures demand variability relative to the average:
CV = Standard Deviation of Demand / Mean Demand
Where:
- Standard Deviation: How spread out your weekly (or monthly) sales quantities are from the average.
- Mean Demand: Average units sold per period over your measurement window.
A CV of 0.3 means demand typically varies by 30% from the average. A CV of 1.5 means typical variation exceeds the average itself, which signals erratic behavior.
The ABC-XYZ Matrix
| X (CV < 0.5) | Y (CV 0.5-1.0) | Z (CV > 1.0) | |
|---|---|---|---|
| A | AX: Auto-replenish, tight safety stock, 97% SL | AY: Forecast with buffer, 95% SL, biweekly review | AZ: Large safety buffer, manual review, 95% SL |
| B | BX: Standard reorder, moderate safety stock | BY: Forecast with caution, wider buffers | BZ: Conservative stock, watch closely |
| C | CX: Min-max reorder, low priority | CY: Periodic review, question catalog fit | CZ: Discontinuation candidate, minimal investment |
Worked Example: Classifying Three FBA SKUs
A $3M FBA seller pulls 26 weeks of unit sales for three products:
SKU A: Silicone Kitchen Tongs ($28 ASP)
Weekly sales: 85, 92, 88, 79, 91, 86, 83, 90, 87, 84, 93, 88, 82, 89, 91, 85, 87, 90, 86, 83, 88, 92, 84, 87, 89, 86
Mean = 87.1 | Std Dev = 3.5 | CV = 0.04 → X
SKU B: Seasonal Grill Brush ($42 ASP)
Weekly sales: 22, 28, 35, 48, 62, 71, 85, 91, 78, 65, 52, 38, 30, 25, 22, 29, 36, 50, 64, 73, 86, 89, 76, 61, 48, 33
Mean = 52.6 | Std Dev = 22.4 | CV = 0.43 → X (borderline Y)
SKU C: Specialty Knife Sharpener ($67 ASP)
Weekly sales: 0, 0, 3, 0, 0, 0, 12, 0, 0, 1, 0, 0, 0, 8, 0, 0, 0, 0, 5, 0, 0, 0, 0, 15, 0, 0
Mean = 1.7 | Std Dev = 4.2 | CV = 2.47 → Z
If SKU A is also an A-tier revenue item, it falls in the AX cell: auto-replenish with a 60-day lead time order triggered at the reorder point, 97% service level, minimal safety stock (demand is predictable). SKU C, even if it generates decent revenue per sale at $67, falls in Z territory and needs Croston's method for forecasting with a much larger proportional safety buffer.
Why Demand Predictability Matters More on Amazon
Amazon's marketplace amplifies demand variability in ways traditional retail does not. Algorithm changes can shift your organic ranking overnight. A competitor stocking out sends their traffic to your listing for a week, then it vanishes when they restock. Lightning deals and Prime Day create artificial spikes. PPC campaign changes alter velocity immediately.
All of this means FBA sellers typically have a higher proportion of Y and Z items than brick-and-mortar retailers. Your demand forecasting approach must account for this. Using a single forecasting method across your entire catalog treats an AX product the same as a CZ product, which wastes accuracy on stable items and gives false confidence on erratic ones.
FBA receiving delays (often 1-3 weeks) compound the problem. If a Z-item suddenly spikes and you reorder, that inventory will not be available for 75-100 days (lead time + receiving). By then, the spike may be over. For Z items, the question is not just "how much to order" but "whether to chase the spike at all."
Common XYZ Analysis Mistakes
1. Using too short a data window. Calculating CV from 4-6 weeks of data produces unreliable classifications. A single promotional week can make a stable SKU look erratic. Use at least 26 weeks (6 months), ideally 52 weeks to capture full seasonality. Short windows cause constant tier-switching that destabilizes your reorder policies.
2. Ignoring seasonality in the CV calculation. A product with strong seasonal patterns (pool accessories, holiday decor) will show a high CV over 12 months even though its demand is predictable. Deseasonalize the data before calculating CV, or calculate CV within each season separately. Otherwise, you will classify predictably seasonal items as Z when they are actually Y or even X within their selling season.
3. Treating all Z items the same way. A $67 specialty item selling 44 units per month erratically (AZ) needs a large safety buffer and close monitoring. A $22 accessory selling 3 units per month erratically (CZ) is a removal candidate. The Z classification tells you the demand is unpredictable; the ABC tier tells you whether it is worth the effort to manage that unpredictability.
Related Glossary Terms
Frequently Asked Questions
How do I combine ABC and XYZ analysis?
Create a 9-cell matrix by crossing ABC tiers (revenue contribution) with XYZ tiers (demand predictability). Each cell gets its own inventory policy: reorder method, safety stock formula, service level target, and review frequency. AX items get automated replenishment with tight buffers. CZ items get minimal investment or removal consideration.
What forecasting method works best for Z items?
Standard time-series methods (moving average, exponential smoothing) fail on Z items because demand is too erratic. Use Croston's method or the Syntetos-Boylan Approximation (SBA), which separate the demand interval from the demand size. For high-value AZ items, supplement quantitative methods with qualitative inputs like upcoming promotions and competitor monitoring.
How much historical data do I need for XYZ analysis?
Use at least 26 weeks (6 months) of weekly sales data, ideally 52 weeks. Shorter windows produce unreliable CV calculations because a single promotional spike or stockout can skew the result. If a SKU has fewer than 13 weeks of history, flag it as unclassified rather than forcing a category.
Should I recalculate XYZ tiers every month?
Quarterly is the sweet spot for most catalogs. Monthly recalculation causes excessive tier-switching that destabilizes your reorder policies. The exception is if you run frequent promotions or deals that significantly shift demand patterns between periods.
What CV thresholds should I use for X, Y, and Z?
Common starting thresholds are X below 0.5, Y between 0.5 and 1.0, and Z above 1.0. Adjust based on your catalog distribution. If 80% of SKUs land in Z, your thresholds are too tight for your product mix. Aim for roughly 20-30% in each bucket.