What Is Lumpy Demand in FBA Inventory?
Lumpy demand is the most challenging demand pattern an FBA seller can face. It combines two sources of unpredictability: sales events are infrequent (separated by zero-sales periods), and the quantity sold when demand does appear varies wildly from one occurrence to the next.
In the Syntetos-Boylan classification framework, lumpy demand occupies the high-ADI, high-CV² quadrant:
- ADI (Average Demand Interval) > 1.32: Long gaps between sales events.
- CV² of non-zero demand sizes ≥ 0.49: When sales occur, the quantity is inconsistent.
Compare this to sporadic demand, where sales are also infrequent but the order quantity is relatively stable. With lumpy demand, you cannot predict when the next sale will happen or how large it will be. A SKU might sell 3 units one week, nothing for a month, then 45 units in a single day. Standard demand forecasting methods collapse under these conditions.
The Syntetos-Boylan Demand Classification
Lumpy demand is one of four patterns in the framework. Understanding all four helps you assign the right forecasting method to each SKU:
| Pattern | ADI | CV² | Description | Forecast Method |
|---|---|---|---|---|
| Smooth | ≤ 1.32 | < 0.49 | Frequent, consistent sales | Exponential smoothing |
| Erratic | ≤ 1.32 | ≥ 0.49 | Frequent, variable sales | Weighted moving average |
| Sporadic | > 1.32 | < 0.49 | Infrequent, consistent sales | Croston's method |
| Lumpy | > 1.32 | ≥ 0.49 | Infrequent, variable sales | SBA (Syntetos-Boylan Approximation) |
The SBA modifies Croston's method by applying a bias correction factor that accounts for the overestimation inherent in standard Croston's when demand sizes vary significantly. For lumpy demand items, this correction matters: it can reduce forecast error by 10-20% compared to uncorrected Croston's.
Worked Example: Lumpy Demand for an FBA SKU
A seller carries a $65 specialty grilling accessory with a 75-day lead time from Ningbo. Here are 26 weeks of unit sales:
0, 0, 3, 0, 0, 0, 0, 28, 0, 0, 0, 5, 0, 0, 0, 0, 0, 45, 0, 0, 8, 0, 0, 0, 0, 12
Step 1: Calculate ADI. Non-zero weeks: 6 out of 26. ADI = 26/6 = 4.33. Well above 1.32.
Step 2: Calculate CV². Non-zero quantities: 3, 28, 5, 45, 8, 12. Mean = 16.83, Std Dev = 16.29. CV² = (16.29/16.83)² = 0.94. Above 0.49. Classification: lumpy demand.
Step 3: Apply SBA. Croston's demand size estimate: ~16.83 units. Croston's interval estimate: ~4.33 weeks. SBA correction factor: multiply by (1 - α/2) where α is the smoothing constant (typically 0.1-0.2). SBA-adjusted weekly demand ≈ 3.5 units/week.
Step 4: Lead time demand. 75 days = ~10.7 weeks. Lead time demand = 3.5 x 10.7 = 37.5 units. Safety stock for lumpy demand needs a compound Poisson approach rather than normal distribution: approximately 30 additional units at 95% service level (much higher proportionally than a smooth-demand SKU).
Reorder point: 68 units. Reorder quantity: 101 units (covers ~29 weeks).
The 45-unit spike in week 18 drives much of the variance. If that was an Amazon Business B2B order, future spikes of that size are plausible. If it was a Lightning Deal, it may not recur and could be filtered from the training data.
What Creates Lumpy Demand on Amazon
Lumpy demand is especially dangerous for FBA sellers because of the long replenishment pipeline. With 45-90 day lead times from Asian suppliers plus 1-3 weeks of Amazon receiving, you cannot react to a sudden demand spike. The inventory you have on hand is all you get for the next 2-3 months.
Several Amazon-specific factors create lumps:
- Amazon Business (B2B) orders: A single business buyer can order 50+ units at once, creating a massive spike that never repeats.
- Lightning Deals and coupons: Promotional events compress weeks of demand into hours.
- Competitor stockouts: When a competitor runs out, their traffic temporarily floods your listing. When they restock, it vanishes.
- Seasonal gift buying: Some products see 10x volume in December then near-zero in January.
Storage limits make it risky to hold the large safety buffers that lumpy items need. This creates a tension: the SKU needs more buffer than average, but Amazon restricts how much you can store. Hybrid fulfillment (small FBA buffer + 3PL backup) is often the practical solution for high-value lumpy items.
Common Lumpy Demand Mistakes
1. Averaging lumpy demand and calling it a forecast. Taking total units sold over 26 weeks (101) and dividing by 26 gives 3.88 units/week. That number is technically correct but practically useless. It hides the fact that 73% of weeks had zero sales and one week had 45 units. An average-based reorder point would leave you stocked out during spikes and overstocked during dead periods.
2. Using the same safety stock formula as smooth-demand SKUs. Normal-distribution-based safety stock assumes demand follows a bell curve. Lumpy demand does not. It follows something closer to a compound Poisson distribution with heavy right-tail probability. Applying a normal z-score to lumpy demand understates the buffer you need by 30-50%, leading to repeated stockouts during demand bursts.
3. Panic-ordering after a spike. A 45-unit week feels like a signal that demand is accelerating. It usually is not. Placing a rush order based on one spike leads to excess inventory that sits through the next 4-6 weeks of zeros. Before reacting, check whether the spike was a one-time event (deal, B2B order) or a genuine shift in the demand pattern.
Related Glossary Terms
Frequently Asked Questions
What causes lumpy demand on Amazon?
Common causes include B2B orders through Amazon Business (a single buyer orders 50 units at once), Lightning Deals creating artificial spikes, competitor stockouts sending temporary traffic your way, and seasonal gift-buying patterns. Any event that creates large, unpredictable order clusters produces lumpy demand.
How do I forecast lumpy demand?
Use the Syntetos-Boylan Approximation (SBA), a modified version of Croston's method that corrects for upward bias. SBA separates the forecast into demand size and interval components, then applies a correction factor. Standard moving averages and exponential smoothing will fail on lumpy patterns.
Should I keep lumpy demand SKUs in my catalog?
It depends on revenue and margin when sales occur. A lumpy SKU generating $3,000 in a single burst sale with 30% margin is worth managing despite the forecasting difficulty. A lumpy SKU generating $80 per sporadic sale is probably not worth the storage and management overhead. Run the sell-through math.
How much safety stock do lumpy demand items need?
More than smooth or sporadic items at the same service level. Because both timing and size of demand are unpredictable, safety stock must buffer against two sources of uncertainty simultaneously. Use a compound Poisson distribution or bootstrap simulation rather than normal-distribution formulas.
What is the difference between lumpy and erratic demand?
Erratic demand sells frequently but in variable quantities (low ADI, high CV²). Lumpy demand sells infrequently AND in variable quantities (high ADI, high CV²). Both have unpredictable order sizes, but lumpy adds the extra challenge of unpredictable timing between sales events.