Divide units sold by the days the product had stock, not by the days in the window, and compare several windows before trusting one. A 30-day window with nine days out of stock reads 30% low. A 7-day window that held a promotion can read double. The rate you reorder from is a judgment you make from both.
A day with nothing on the shelf did not measure demand. It measured the shelf.
Only the days with stock measured demand
A product sold 10 units a day for 21 days, then ran out and sat at zero for 9. Over the 30-day window it sold 210 units.
The arithmetic
By calendar days = 210 ÷ 30 = 7 a day, 49 a week
By in-stock days = 210 ÷ 21 = 10 a day, 70 a week
Nothing about demand changed on day 22. Customers kept looking; there was nothing to buy. Dividing by 30 spreads 21 days of sales over 30 and reports a product that slowed down by 30%.
What the low rate does to the next order
Take an order meant to cover 16 weeks: production, freight, the gap until the next review and a buffer. The reorder quantity formula multiplies the rate by those weeks:
- At 49 a week: 784 units.
- At 70 a week: 1,120 units.
The calendar-day rate orders 336 units too few, nearly five weeks of real sales. Those five weeks come out of the buffer first and then out of the shelf, and the next stockout puts more zero days into the next window, which lowers the next rate again. A stockout that is not corrected for teaches the spreadsheet to order less each time.
When a window is too thin to use
The shorter windows of the same product carry almost no evidence. Its last 14 days held 5 days with stock, and its last 7 days held none. A 7-day rate of zero measures a window in which the product could not sell. A 14-day rate built on 5 days is a measurement of 5 days.
Vendlen's selling rate applies two rules here that a spreadsheet can copy. A day counts as out of stock when sellable stock was below one week of the applied rate, not only when it was at zero, because a nearly empty shelf may not have measured demand either. And a window with fewer than half its days in stock is skipped and listed as such, so the longer windows carry the rate instead.
Five windows, read as a shape
A different product, in stock throughout, sold 10 a day until a promotion last week sold 20 a day. Each window converted to a weekly figure:
| Window | Units sold | Per week |
|---|---|---|
| 7 days | 140 | 140.0 |
| 14 days | 210 | 105.0 |
| 30 days | 370 | 86.3 |
| 60 days | 670 | 78.2 |
| 90 days | 970 | 75.4 |
Five windows that agree describe a settled product. Here the short windows sit above the long ones, falling steadily, which says something changed recently. The windows cannot say what. You can: you ran the promotion.
Reordering 16 weeks at the 7-day rate buys 2,240 units; at the underlying 70 a week, 1,120. The extra 1,120 units are 16 weeks of normal sales bought because of one good week. Now suppose there was no promotion and 140 a week is the new normal, because a competitor left or a listing change worked. Then the 90-day rate is the wrong one, and 16 weeks at 75.4 orders about 1,207 units when 2,240 were needed. The table looks the same in both stories. Which one is true is the part only you know.
How far a rate should move in a week
Even a correct rate should not jump to every new reading. Vendlen holds the applied rate inside a floor and a ceiling around the previous week's rate, and the band is not symmetrical: the rate rises faster than it falls, because being short costs more than holding a little too much.
A stockout gets its own rule, because there is no demand to measure while it lasts. The rate is held for the first 14 days, then steps down 10% a week, and stops at half the rate before the stockout. For the first product, which sold 70 a week before it ran out, that floor is 35 a week, and the rate stays there until stock is back. A spreadsheet that lets the rate fall to zero during a stockout will propose no order at the moment one is most needed.
Where this still needs a person
Every window is the past. A promotion you are about to run, a seasonal peak, a launch, or a listing that was suppressed for half the window are things you know and the data does not. Override the rate for those, write down why, and expect to revisit it. In Vendlen an override applies to that SKU and channel only, and order proposals pause until the rate is recalculated around it.
A missing rate is also not zero demand. A product with no usable sales history needs a rate set by hand or no proposal at all; Vendlen shows it as missing data rather than planning to zero. The same averaging problem turns up in advertising data, where one long window hides a fortnight that broke. For the rest of the inventory chain, see Inventory and Supply.
What to do with this
- Pull 90 days of daily units sold and daily stock for the SKU, channel by channel.
- Mark the days with no stock, and the days the shelf was nearly empty. Leave both out of the units and the days.
- Work the 7, 14, 30, 60 and 90-day rates as units per in-stock day, times 7. Drop any window with fewer than half its days in stock.
- Read the shape. If the windows agree, use the rate. If the short ones sit above the long ones, or below, find the reason before you pick.
- Write the rate you chose and the reason beside the order, so the next review can check it.
Questions sellers ask
What is sales velocity on Amazon?
The number of units one SKU sells per day or per week on one channel. It drives days of supply and every reorder, so an error in it carries straight into the order. Measure it on days the product was in stock.
Should I use the 30-day or the 90-day sales average?
Neither on its own. Work out several windows on in-stock days and compare them. If they agree, the rate is settled; if the short windows sit well above or below the long ones, something changed recently, and you need to know what before choosing.
How do I handle a promotion when calculating sales velocity?
Leave the promotion days out of the rate you reorder from, or use windows long enough that one week barely moves them, and plan the promotion's own units as a separate, known event. Reordering many weeks of cover at a promotion week's rate buys stock for demand that has already ended.
Why does my sales velocity drop after a stockout?
Because the days at zero stay in the window. A calculation that divides by calendar days counts them as days the product did not sell, when it could not sell. Divide by in-stock days. During a long stockout, hold the last good rate for a while and let it step down slowly, rather than letting it fall to zero.