How to Model Amazon Sales Seasonality for FBA Sellers | Inventory Hero
·6 min readForecasting
How to Model Amazon Sales Seasonality for FBA Sellers
How to model Amazon sales seasonality: build seasonal indices from a year of history, forecast a baseline, and apply the index ahead of your lead time.
Take at least a year of sales history, find the average sales per period, then divide each period's actual sales by that average. The result is the seasonal index for that period: 1.0 is an average period, 1.5 is 50 percent above average, 0.6 is 40 percent below. Apply the index by forecasting a baseline and multiplying by the relevant period's index.
How much history do I need to model seasonality?
At least one full year, and two is much better because it lets you separate a real seasonal pattern from a one-off event. Exclude any periods the SKU was out of stock, or a past stockout will look like a low season and pull that period's index down incorrectly.
T. Brian Jones is co-founder and CTO of Inventory Hero. He leads the engineering behind its Amazon data pipeline, demand forecasting, and the AI platform that lets sellers talk to their live inventory, sales, and supplier data in plain language.
Start a full lead time before the demand ramp begins. If your lead time is 80 days and your peak is late November, the order has to be placed in late summer. Waiting until sales climb means the stock lands after the season has already started.
Read article
To model Amazon sales seasonality, build a seasonal index from at least a year of history, forecast a baseline with your usual method, then multiply the baseline by the index for the period you are forecasting. The short version: a December index of 1.8 means December runs 80 percent above your average month, so you scale the forecast up accordingly, and you do it a full lead time before the ramp. Below is how to build the indices, how much history you need, and the FBA-specific timing that makes seasonal SKUs unforgiving.
A seasonal index is each period's typical sales expressed relative to the yearly average. You compute it in three steps:
Take at least a year of sales history per SKU.
Find the average sales per period (per month, or per week for finer control).
Divide each period's actual sales by that average.
The result is a multiplier. An index of 1.0 is an average period; 1.8 means that period runs 80 percent above average; 0.6 means 40 percent below. For a SKU with a strong Q4, the indices might run near 0.7 in the slow spring months and climb past 1.5 in November and December.
Say a SKU averages 300 units a month across the year, and last November it sold 540:
Step
Value
Yearly average per month
300 units
November actual
540 units
November index
540 / 300 = 1.8
Now, to forecast next November, you forecast the baseline (say your trend says the SKU will average 360 a month next year) and multiply by 1.8: 360 x 1.8 = 648 units for November. The baseline captures the growth; the index captures the season. For the baseline methods, see inventory forecasting methods.
One full year is the minimum, because you cannot see a yearly pattern in less than a year. Two years is much better, because it lets you tell a real, repeating season from a one-time event: a spike that happened once is noise, a spike that happened in the same month two years running is seasonality.
Critically, exclude out-of-stock days from the history. If the SKU stocked out last December, that month's sales read artificially low, and a naive index would tell you December is a slow month, exactly backward. This is the same out-of-stock correction you make for sales velocity, and it matters even more here because one missed peak distorts the whole pattern. To do it: pull the units-ordered history from your Seller Central Business Reports, then open the Inventory Ledger report (under the FBA inventory reports), which shows your daily ending sellable quantity, find the days the SKU hit zero, and drop those days from the period before you compute its index.
Modeling the season is the easy part. The hard part is that on FBA you have to act on it a full lead time early. If your peak is late November and your total lead time is 80 days, the purchase order has to go out in late summer, long before sales give you any sign the season is coming. Waiting for velocity to climb guarantees the stock lands after the ramp has started.
Three FBA constraints make the timing even tighter:
Q4 storage surcharges. Holding inventory through the fourth quarter costs more per cubic foot, so overbuilding a seasonal SKU is more expensive than usual.1
Capacity and restock limits. The window when you most need to send in stock is when limits are tightest, so a seasonal build can hit the ceiling. See restock limits.
Slower check-in. Amazon check-in stretches during peak, adding weeks to the leg you control least.
Plan the build backward from the peak through the full lead time; the reorder point calculator helps you set that order date from the peak minus your lead time. Then plan the drawdown, which sellers do worse than the build. Say you built to that 648-unit November peak and December still runs hot at an index of 1.5 (about 450 units). You want to enter January nearly empty, because at a post-season rate of maybe 8 units a day you would clear only around 250 units in January, and anything much above that becomes dead stock paying the higher winter storage rate. Ease the price down through December rather than dumping it in January, so the SKU exits the season near zero instead of carrying inventory into the new year.
Using one year and trusting it. One December does not prove a pattern. Confirm with a second year where you can.
Letting a past stockout depress an index. Exclude out-of-stock days or your peak month reads as a trough.
Building too late. The most common and most expensive seasonal error: ordering when sales rise instead of a lead time ahead.
Forgetting the drawdown. Stock you build for the peak has to sell through, or it becomes January dead stock paying surcharges.
Letting a promotion distort the index. A deal week or coupon inflates that period's sales, so it reads as a higher season than it really is. Strip out promotion-driven spikes before you build indices, the same way you treat events separately in a baseline forecast.
Modeling Amazon sales seasonality means building a seasonal index from a year or two of clean history, forecasting a baseline, and multiplying by the index, then acting on it a full lead time before the peak. Keep the baseline and the season as separate steps, exclude out-of-stock days, and plan both the build and the drawdown. For the wider system, see Amazon inventory forecasting and how to forecast Q4 inventory.