Forecasting

Sales are declining, and why the reorder math still says buy the same amount

Every cycle it buys a little more of something that is selling a little less, and nothing on the screen calls that a mistake.

Marcus VolstedCo-founder, Restocio Published 10 October 202611 min read

A trailing sales rate is not wrong on a product that is declining. It is late. The days behind you always include some of the old, faster pace, so the number an order gets sized on sits above where the product actually is right now. That is not a flaw in one particular app. It is what an average does by construction, in a spreadsheet or in a purchasing tool, and it is the mechanism that turns a product that is merely slowing down into the dead stock at the back of the warehouse.

Two other posts on this site cover the edges of a product's life. Forecasting a new product is the beginning, where there is not yet enough history to trust. Dead stock is the end, where the aging bands take over because the money is mostly already spent. Nothing here has covered the middle, which is where the actual damage happens: a product with months of ordinary sales history that has quietly started selling less than it did a season ago, while the reorder math keeps working from the pace it used to keep.

Why a trailing rate always lags a falling product

A sales rate measured over the last 30, 60 or 90 days is a blend. While a product's pace is flat, the blend and the truth are the same number, and the window length barely matters. The moment the pace actually changes, every day still inside that window that predates the change keeps voting on today's order, and the wider the window, the longer the old pace keeps a say.

Put a number on it. A product sells a steady 10 units a day for months, then genuinely drops to 6 a day, starting 20 days ago. Measure the rate today with three different windows.

Window	What it blends	Rate it reports today	Above the true current pace of 6
30 days	10 days at 10, 20 days at 6	7.3 a day	+22%
60 days	40 days at 10, 20 days at 6	8.7 a day	+44%
90 days	70 days at 10, 20 days at 6	9.1 a day	+52%

Every one of those numbers is a real, honestly computed sales rate. None of them is a mistake in the arithmetic. The 60-day figure is simply describing a product that, twenty days ago, stopped existing in the form the number describes.

The error does not stay stuck at the rate. It multiplies straight into the order, because the order is the rate times however many days of cover the plan is aiming for. Size an order against a coverage target of, say, 150 days at the 60-day rate above and you are buying for 150 × 8.7, about 1,305 units. The same 150 days at the true current pace of 6 a day needs 900. The extra 405 units are not a rounding error. At the true pace they are roughly 68 more days of cover than the plan called for, sitting on a shelf that is already moving slower than the day the order went in.

Where that error goes if nobody catches it

It does not correct itself in one step. Next cycle, the window has rolled forward, so it holds fewer of the old fast days and more of the new slow ones, and the reported rate edges closer to the truth. But the order already placed does not shrink to match. It sits on the shelf at its full size, days of cover creeps up rather than down, and the ordinary read on rising days of cover, that the product is comfortably stocked, is the same read a healthy product gives. ABC analysis makes a related point from a different angle: a class letter assigned from last year's value can stay the same for a full quarter while the product's direction changes underneath it. A trailing rate has the same blind spot, just measured in weeks instead of months. Repeat the cycle two or three times on a product that keeps slipping and the gap between what was bought and what actually sells is exactly where dead stock comes from, not from one bad decision but from an ordinary calculation running honestly on a number that stopped being true weeks earlier.

Is this a decline, or a slow fortnight?

A single low window proves very little on its own, and reacting to it is its own mistake, covered below. Before trimming an order, it is worth checking the fall against a few things that are cheap to look at.

Rule out a stockout first. A fall in the sales rate that lines up with the product actually running out is a supply story, not a demand one, and belongs with the stockout correction this site argues for everywhere else, not with a genuine decline. If the shelf stayed full the whole time, that explanation is off the table.

Look for two windows in a row, not one. One slow 30-day period against a strong prior one is inside the range that ordinary week-to-week noise produces on almost any product. Two consecutive periods below the one before them is a much harder pattern to explain away as luck, and it is the simplest trigger that does not need a statistics course to apply.

Ask what changed, because something usually did. A price increase, a competitor's listing going live, a review score that dropped, an ad account that got shut down, a bestseller badge that moved to a different product. None of that shows up in a sales number by itself, but a person who watches the store usually knows within a sentence or two whether there is a story behind the fall.

Check the same weeks last year, if you have them. A full year of history rules out an ordinary seasonal dip; anything less than that cannot rule it out, and no arithmetic changes that. The seasonality standard the statistics agencies use is stricter than most stores realise, and it is worth reading before trusting a pattern built on a handful of months.

	A normal seasonal dip	A product that is actually declining
Same weeks last year	Shows the same dip	Shows no matching dip, or a smaller one
Coincides with a stockout	No, the shelf stayed full	Sometimes, so check in-stock days before trusting it
What the next window usually does	Recovers on its own once the season turns	Holds at the lower pace or keeps falling
Other products in the same category	Move the same way, together	Stay flat. It is isolated to this one
Right response	Keep the plan, the dip is expected	Shrink the next cycle and re-check sooner

What to change about the order, not the product

This is a forecasting post, not a clearance one, so the actions here stop at the reorder decision. What to do with stock you already bought too much of is a different question with its own answer.

Forecasting a new product makes an argument that applies here in reverse. On a product with too little history to trust, the right lever is the cycle length, how long a single order is meant to last, because the pipeline and cushion days protect the shelf from a late delivery and should not be touched. A product with plenty of history and a recent break in it is the same problem from the other side: plenty of data, just a data point you no longer trust. The same lever applies. Shrink the cycle so less of the order rides on a rate you suspect is stale, and leave the pipeline and cushion days doing the job they were sized for, which is surviving a late delivery, not guessing at direction.

Re-measuring over a shorter window helps too, on purpose rather than by accident. A 30-day window reacts to a real shift faster than a 60 or 90-day one, at the cost of being noisier when nothing has actually changed, a trade-off explained in full on the forecasting pillar. It is worth making deliberately on the one product you are watching, not across the whole catalogue, where the wider window is doing its job well.

What not to do is treat the first slow window as confirmed and cut the order to nothing. A product that recovers after a genuinely quiet fortnight and finds its shelf half empty has traded a smaller mistake for a real one. The point of checking for noise first is exactly to avoid this.

What this looks like inside Restocio

The number that sizes an order is a trailing sales rate measured over the days a product was actually in stock, over a 60-day window by default. It carries no trend and no direction. A least-squares trend line exists on the Demand page for the shop as a whole, and it is shown there with its measured error against a flat average, but it does not resize any individual product's order, so it will not notice this on your behalf. One case where the trailing rate is set aside is a product marked seasonal, in the weeks before its season starts, where last year's same months are used instead. That solves a different problem, a product with nothing current to measure yet, not a product whose current pace has genuinely changed, and it is covered in full on the forecasting pillar linked above.

The override that helps a new product get a shorter first cycle works the same way here. Type a quantity directly on a freight line for the one product you are watching, and that number replaces the app's figure for that line while every other line on the plan keeps running on the app's own arithmetic. It is manual, and it is meant to be, because deciding that a fall is real rather than noise is exactly the judgement call the app should not make quietly on its own. The rest of the loop, from the plan to the purchase order, is in the Shopify replenishment guide.

How sure can you actually be with a few weeks of data

Not very, and this page will not pretend otherwise. Formally separating a real change in direction from ordinary noise needs more observations than a typical reorder cycle allows you to wait for, and a store that only checks a product every 60 or 90 days will always be acting on a handful of data points, not a chart a statistician would sign off on. The checks above, two consecutive weak windows, a stockout ruled out, a look at the same weeks last year, are a practical filter, not a hypothesis test, and they will occasionally be wrong in both directions: a real decline that still looks like noise for one more cycle, and a genuine slow patch that gets its cycle trimmed for nothing.

That is the honest tradeoff, and it runs the same way whether the arithmetic behind it lives in an app or a spreadsheet. Waiting for certainty costs the weeks a real decline keeps getting bought at the old rate. Acting early costs an occasional shrunk order on a product that was fine. Between those two, the second mistake is the cheaper one to make repeatedly, which is the whole argument for shrinking the cycle rather than guessing at the product's fate.

Common questions

Sales are declining. Should I still reorder the same amount?

Not automatically. Check first whether the fall lines up with a stockout, whether it shows up in two consecutive measuring windows rather than one, and whether the same weeks last year looked the same. If the decline holds up against those checks, shrink the cycle length on the next order rather than the pipeline or cushion days, so less of the order rides on a rate you no longer trust.

How can I tell if a product is actually dying, or just having a slow couple of weeks?

One weak window rarely proves anything on its own. Look for two consecutive periods below the one before them, confirm the shelf stayed stocked the whole time so it is not a supply problem wearing a demand costume, and compare against the same weeks last year if you have that much history. None of that is certainty, but it separates a pattern from a single bad week.

Why does my reorder tool keep telling me to buy the same amount when sales are obviously falling?

Because a trailing sales rate is a backward-looking average, and while a product's pace is changing, the average always sits somewhere between the old pace and the new one. That is true of any velocity-based reorder math, a spreadsheet formula or an app, not a flaw specific to one tool. A shorter measuring window reacts faster, at the cost of being noisier when nothing has actually changed.

Is a declining sales rate the same thing as dead stock?

No, they are different points on the same line. A declining rate is the mechanism, the reorder math sizing an order slightly too large every cycle because it is working from a pace the product no longer keeps. Dead stock is what that mechanism produces if nobody catches it, stock sitting on the shelf well past the point where it is worth what was paid for it.

How many weeks of falling sales should I wait before acting?

There is no universal number, and treating this page as one would be dishonest about how thin a few weeks of data really is. Two consecutive measuring windows lower than the one before them, with a stockout ruled out, is a reasonable trigger to shrink the next cycle. It is a practical filter, not a statistical test, and it will occasionally be wrong in both directions.

Marcus Volsted

Co-founder, Restocio

Marcus co-founded Restocio and works on it daily with a Swedish importer who plans their purchasing in it every working day. Restocio is built in Sweden by two founders, one Swedish and one Danish, and Marcus is the Danish one. Why we are building it.