- A large assortment cut at an online grocer lowered overall store sales, and most of that loss came from people shopping less often rather than buying less per visit.
- Rank by cash margin after discounts and returns, not by revenue. Returns ran at 15.8% of US retail sales in 2025 and 19.3% online, so a high return product can rank well on revenue and lose money.
- US retailers held 1.36 months of inventory for every month of sales in August 2026. If your own ratio is far above that, the problem is breadth rather than demand.
- Stock that will not sell at cost has to be carried at net realisable value under IAS 2, so a write down is an accounting duty rather than an admission of defeat.
- Move a product before you kill it. Placement and presentation explain a surprising share of what looks like dead demand.
Every catalogue grows the same way. A supplier offers a colour you did not plan to stock. A customer asks for a size twice. Something sells well, so you add the variant next to it. Nobody ever holds a meeting about adding one product, and nobody ever holds a meeting about the eighty you now carry that each sold four units last year.
The instinct, once you notice, is to cut hard. Pull the bottom quarter, free the cash, simplify the photography. That instinct is reasonable and the evidence says it is also risky, because the products that look worst on a sales report are not always the ones costing you money.
Why does cutting the tail sometimes cut sales?
Because customers do not shop item by item. When an online grocer made a large reduction to its assortment, researchers found overall store sales fell, and the bigger part of the loss came from shoppers visiting less often rather than spending less per visit.
That study, published in Marketing Science in 2005 by Borle and colleagues, modelled store level and category level effects at the same time across every category in the store. Its result ran against the earlier literature, which had generally found assortment cuts to be harmless or mildly positive. The detail that matters for a small shop is where the damage concentrated: less frequently purchased categories were hurt more, and the authors suggest comparing categories rather than applying one rule across the whole range.
Translate that into a shop of one or two people. The risk is not that you delete a product nobody wanted. It is that you delete the reason somebody opened your site at all. A slow moving item can be the thing a customer searches for, finds, and then buys three fast moving items alongside. Delete it and the report will show you saved a dead SKU. It will not show you the visit that stopped happening.
Which number should rank your products?
Cash margin after discounts and returns, divided by the stock it ties up. Revenue rankings are the default in every dashboard and they are the wrong tool, because they ignore both the money you gave away to sell the thing and the money it costs you when it comes back.
Build the column yourself once. Take the unit's selling price, subtract the discount actually realised rather than the list price, subtract landed cost, subtract the payment fee, then subtract the return cost multiplied by that product's own return rate. What remains is the cash each sale contributes. Divide the annual total by the average stock value held and you have the number that tells you whether a product earns its shelf.
Doing this by hand for 300 products is unpleasant, which is exactly why it is worth automating once rather than repeating by instinct every season. The groundwork is the same groundwork any forecast needs, and we set it out in what an inventory forecast requires before it can help you. Clean cost data first, cleverness second.
| Signal | What it looks like in the data | What it usually means | Reasonable action |
|---|---|---|---|
| Slow but profitable | Few units, healthy cash margin, low stock held | A niche line that costs almost nothing to carry | Keep, order smaller quantities |
| Fast but thin | High revenue rank, margin collapses after discounts | You are buying sales with price | Reprice or stop promoting it |
| High return rate | Return rate well above your catalogue average | A listing problem or a product problem | Fix the page before cutting the line |
| Basket opener | Low direct sales, often first item in the order | It brings the visit that carries the rest | Keep regardless of its own rank |
| Quietly dead | No sales, stock held, no basket role | Capital sitting still | Exit, and write it down properly |
What do returns do to the ranking?
They reorder it. The National Retail Federation put total US returns at $849.9 billion in 2025, which is 15.8% of annual retail sales, and 19.3% of online sales. A catalogue average hides enormous variation between products, and the variation is where the money is.
Apply the arithmetic. A product selling 100 units a year at 12 pounds of gross margin looks like 1,200 pounds. Give it a 30% return rate and a return cost of 9 pounds, and you lose 270 pounds in handling before you count the units that come back unsellable. The same shop's 20% margin product with a 4% return rate quietly outperforms it while sitting lower on every revenue report.
The NRF figures carry two further details worth keeping in view. Nine percent of all returns are fraudulent, and close to two thirds of consumers admit to at least one costly returns behaviour. That is a reason to measure return rates per product rather than per customer, because an unusual rate on one line is more often a sizing or photography failure than a crime wave. We went through the operational side of that in who decides on a refund when one order in five comes back.
How much stock is too much?
Compare yourself to the published benchmark before trusting your gut. US retailers held an inventories to sales ratio of 1.36 in August 2026, meaning roughly 1.36 months of merchandise on hand for each month of sales, according to the Census series published by FRED.
That number is an average across very different businesses, so treat it as a sanity check rather than a target. A seller of perishables should sit far below it. A jeweller may legitimately sit far above. What the comparison is good for is catching the case where you are carrying four months of stock in a category turning once a quarter, which is usually a breadth problem rather than a demand problem. Twelve variants each holding three months of cover is the same money as four variants holding a comfortable one.
There is a second reading of the same figure. The ratio moved very little through 2026, sitting at 1.37 in April and 1.36 in August, which tells you the aggregate is stable and that any sharp move in your own number is yours rather than the market's. If your cover doubled this year, nothing happened to retail. Something happened to your buying.
Run the ratio per category rather than for the shop as a whole. One category is almost always carrying the others, and the aggregate number politely conceals it.
What do you do with the stock you decide to drop?
Price it to move and record it honestly, in that order. Under IAS 2, inventories are measured at the lower of cost and net realisable value, which is the estimated selling price in the ordinary course of business less the costs still needed to complete and sell it.
The practical consequence is one most small sellers avoid for emotional reasons. If a line will only ever sell at 40% off, its carrying value is not what you paid, and the write down is recognised as an expense in the period it occurs. Holding dead stock at cost makes your accounts look better and your decisions worse, because every month you delay the markdown you are also paying to store, insure and count something whose value you have already lost.
Exit routes, in rough order of what they preserve. A bundle with a fast mover recovers the most margin and clears slowly. A deep single markdown clears fastest and teaches your regulars to wait for sales, so time it away from your main season. Wholesale or clearance channels move volume at a price that hurts, which is still better than a shelf. Donation has a tax treatment worth asking your accountant about before you skip it. Our piece on when to mark dead stock down covers the timing question that sits underneath all four.
Which products should survive a cut no matter what they sell?
Four kinds, and none of them rank well on a revenue report. Keeping them is not sentiment, it is a decision about what the range is for.
The basket opener comes first, because its value sits in other products' sales. The credibility item is second: the professional grade option, the largest size, the one serious buyers check for before trusting the rest of your catalogue. Third is anything cheap to hold, where a year of carrying cost is a few pounds and the downside of being out of stock is a customer who stops looking. Fourth is the product whose replacement would cost more than the problem, which covers made to order lines and anything you drop ship, since neither ties up cash in the first place.
The honest test for each is a question rather than a formula. If this disappeared tomorrow, what would the customer who bought it do next? If the answer is buy a different one of yours, cut freely. If the answer is go somewhere that still has it, you are not cutting a product, you are cutting a relationship.
Where can software actually help with this?
In the counting, the grouping and the early warning. Not in the decision, because the decision depends on things no dataset holds: what you want the range to be, what your supplier will do if you drop a line, whether the customer who buys it is also the one who tells other people about you.
Three jobs are worth handing over. The first is assembling the cash margin column, which means joining order lines to costs, discounts, payment fees and return records that usually live in four different places. This is tedious and mechanical, and a model that reads your exports and produces one table per product is doing the work a spreadsheet would do if you had the hours. The second is grouping, because a 300 line catalogue has perhaps fifteen real behaviours in it, and clustering products by how they sell rather than by the category you filed them under routinely reveals that two of your categories behave identically and one contains two different businesses.
The third is prediction on the only metric with a usable signal: return rate. Returns correlate strongly with attributes you already hold, including size range, price band, material and how many photographs the listing carries. A model trained on your own history can flag a new product likely to return above your average before you commit to the second order, which is worth more than any ranking of products you already bought.
What to be sceptical about is demand forecasting on a thin history. A line with eleven sales in a year contains almost no information, and a confident forecast on it is an artefact of the method rather than a fact about your customers. Treat anything below roughly thirty observations as a judgement call dressed up in decimals. The same caution applies to automated cut lists: a model can rank, but the moment it is allowed to delete, somebody will discover in March that the hero product's spare part was removed in January.
Do variants count as separate products?
For stock, yes. For the cut decision, usually not. A size curve is one product with a demand distribution attached, and deleting the tails of it is a different operation from deleting the line.
This distinction causes more quiet damage than any other part of the exercise. The smallest and largest sizes almost always sell least, which puts them at the bottom of every ranking, and dropping them converts a complete offer into an incomplete one. The customer who wears that size does not buy the next size up. They leave, and they tend not to come back once they have learned your range does not cover them, which means the cost lands in future months where nobody attributes it correctly.
Colour and material variants behave differently and can be pruned harder, since a buyer refused black will often take navy. The useful rule is to cut across variants only where a substitute exists inside your own range, and to cut whole lines where it does not. Where a supplier sets a minimum order quantity per variant, that constraint, rather than your sales data, is frequently the real reason a tail exists, and renegotiating the minimum is a cheaper fix than deleting the colour.
How often should you run this?
Once a quarter, on a fixed date, with a written rule. Ad hoc culls happen when cash is tight, which is precisely when the decision gets made badly and the basket openers go first.
A quarterly pass takes an afternoon once the data is in shape. Rank by cash return on stock held, flag anything in the bottom decile, then apply the four exemptions above before anything is deleted. Anything that survives the exemptions goes on a watch list for one more quarter rather than straight to the exit, with a small test attached: move it on the category page, rewrite the title, photograph it against a different background. Placement and presentation explain more apparently dead demand than most sellers expect, and a test costs nothing compared with the cash already sunk in the stock.
Keep the decisions in writing. A one line note saying what was cut and why turns a seasonal purge into a record you can learn from, and it stops the same product being reordered next year by whoever is doing the buying. If you want the catalogue itself to carry that metadata, product status and review dates alongside the price, that is the kind of field a shop you can shape yourself can hold without waiting for a platform to support it.
One more thing before the first cut. Pull the list of what you intend to delete and check it against your own search logs for the last ninety days. Products with no sales but real searches are not dead. They are either out of stock at the moment people want them, priced wrongly, or sitting behind a photo that does not answer the question the searcher is asking. Deleting those is the one mistake in this whole exercise that does not show up in any report, and the research that decides what to buy next depends on the same signal, so losing it costs twice.