Assortment optimization for retail buyers

The tail is longer than anyone admits

Assortment optimization is usually described as a trimming exercise, and the first number everybody reaches for is the count of SKUs that contribute almost nothing. In the center-store set at our 62-store operator, that number is uncomfortable: the bottom three deciles are 30% of the SKU count and 3.6% of the dollars.

Thirty-eight percent of dollars come from the top decile. By decile four the marginal contribution has fallen under 10%, and by decile eight an entire tenth of the range is contributing 1.8%. A buyer looking at this chart for the first time usually concludes that the bottom three deciles should go, and that conclusion is wrong roughly half the time, for reasons that are the whole point of doing this carefully rather than arithmetically.

Why a low-contribution SKU is not automatically a cut

Four things a dollar-contribution ranking cannot see, each of which saves items that the chart condemns.

The item is a destination, not a volume driver. Some low-volume items bring a specific shopper into the store, and losing them loses the trip rather than the line. The gluten-free or allergen-specific item that sells four units a week is carrying a basket, not a shelf position. Check basket attachment before cutting: what else is in the transaction, and does that shopper appear without this item.

The item is under-distributed, not slow. An item authorized in 14 of 62 stores looks terrible chain-wide and may be performing well where it exists. Ranking by total dollars punishes narrow distribution, which is a decision you made, not a verdict the shopper delivered. Rate and reach are different questions, and the velocity, share, and TDP decision tree is the cleanest formalization of why conflating them misleads.

The item is new. A SKU eleven weeks into distribution has not had time to build repeat, and cutting on trailing-52 dollars systematically kills every launch before it can work. Exclude anything inside its trial window, and be explicit about what that window is.

The item is the price tier. An opening-price-point item may be slow and still be the reason the shopper trusts the set's pricing. Removing it can move demand out of the category rather than up to the next tier.

Apply all four filters to the bottom three deciles and the genuinely deletable list is usually about half of what the raw ranking suggested.

Assortment optimization needs a cut line you can defend

Once the protected items are set aside, you need a rule rather than a judgment call per SKU, because judgment per SKU is how range reviews take six weeks and still get relitigated.

The rule that holds up: cut where marginal contribution falls below the cost of carrying the position.

carrying cost per SKU per year =
    (facings x annual $ per facing) + (admin + ordering overhead)

An item that returns less than that number is destroying value even if it is profitable in isolation, because the space it occupies has a real alternative use. This reframes the question productively. It is no longer "is this item bad," which invites defence, but "does this item beat the item that would replace it," which is the actual decision.

Worked on the center-store set: at $288 in annual sales per selling square foot and roughly 0.9 square feet per facing position, a single-facing item that generates under about $260 a year is not covering the space it sits on. That threshold lands in the eighth decile, not the tenth, which means the honest cut is deeper than most range reviews go and shallower than the raw bottom-30%-of-SKUs instinct.

Adding is the half nobody optimizes

Assortment optimization is treated as subtraction, and most of the damage actually happens on the addition side, because additions are approved one at a time against a forecast the supplier supplied.

The test for any proposed item: what shopper need does it serve that the set does not already serve. If the answer names a segment, a price tier, or a dietary attribute genuinely absent from the set, the forecast can be incremental. If the answer is a cheaper version of an item already in position four, the honest forecast is mostly cannibalization, and the item should be evaluated on whether it improves category margin at constant volume.

The line-extension row is where the most range bloat originates. A supplier launching a fourth flavour forecasts it as incremental, the buyer authorizes it as incremental, and it delivers by taking volume from the other three while adding a fourth position to the planogram and a fourth line to every order. Two years of that is how a set that used to have 240 items has 310 and no more sales.

Sequence the review so it converges

Range reviews sprawl because everything is negotiable at once. Fix the order.

Protected items first, using the four filters above, and publish the list before any cutting discussion begins. Then apply the carrying-cost cut line mechanically to everything unprotected. Then, and only then, consider additions against the space the cuts released. Reversing those last two steps is how you end up adding twelve items and cutting four.

One discipline worth enforcing: additions and deletions balance within the space available, not within the SKU count. Twelve small items out and six large ones in is not a net reduction in anything that matters, and SKU count is the metric that makes it look like one.

What to measure after the reset

The reset either worked or it did not, and the honest read is not next month's category sales, which move for a dozen unrelated reasons.

Measure three things at eight weeks. Category dollars and units against a control group of stores that did not reset, if you have one. Lost-item recovery: for each cut SKU, how much of its volume showed up in the rest of the set rather than leaving. And out-of-stock rate on the survivors, because a reset that broadens facings on winners should improve availability, and if it did not, the planogram did not follow the plan.

The lost-item recovery number is the one that teaches you something durable. If you cut twenty items and recovered 80% of their volume elsewhere in the set, the cut was right and you can cut deeper next time. If you recovered 30%, those items were serving a need the rest of the set does not serve, and the cut line is too aggressive for this category.

The politics of a cut list

Assortment optimization fails on politics more often than on analysis. A cut list touches supplier relationships, a category manager's judgment, and in some chains a store manager's local knowledge, and every one of those is a route to relitigating an item that the numbers already settled.

Three things make the list stick.

Publish the rule before the list. If the carrying-cost threshold and the four protective filters are circulated and agreed a week ahead, the conversation about any individual item becomes a conversation about whether it meets a published criterion. Without the rule first, every item is argued on its own merits, and the person with the strongest anecdote wins each time.

Give suppliers the criteria, not the outcome. A supplier told their item is being cut will produce a deck. A supplier told the threshold six weeks earlier will either produce evidence that their item clears it, which is genuinely useful, or plan for the delist. The second conversation is far cheaper for both sides, and it occasionally surfaces basket or distribution data you did not have.

Record the exceptions and revisit them. Some items survive the cut line for reasons that are real but not in the data: a local brand with a following, an item a major account requires. Write down which items were protected and why, and review that list at the next range review. Unrecorded exceptions accumulate silently, and after three cycles the exception list is the reason the tail grew back.

What a good range review actually costs

Chains under-resource range reviews and then run them annually because they are painful, which is exactly backwards: the pain is largely a function of the infrequency. A set reviewed every year has twelve months of accumulated drift to argue about, and the arguments are proportionally larger.

The realistic cost of a well-supported review is a few days of analysis, one meeting to agree the rule, one to work the exception list, and the reset labor. The analysis is the part that collapses when the data is assembled by hand, and it is also the part that determines whether the other meetings are short. A review that opens with a defensible protected list and a mechanical cut line finishes in two meetings. One that opens with a raw ranking takes six weeks and ends in compromise.

Doing this in Scout

The blocking problem is rarely the analysis, it is that the four protective filters each need a different dataset, and assembling them per range review is what makes range reviews slow enough to run annually rather than when the data says they are needed.

Scout holds the item view with the pieces already joined: contribution, carrying stores versus authorized stores, weeks since first sale, basket attachment, and the space each position occupies. The bottom-decile list arrives already flagged for narrow distribution, recent launch, and basket role, so the conversation starts at the genuinely deletable set rather than at a raw ranking that half the room can immediately object to.

The carrying-cost threshold is computed from the department's own sales per square foot rather than a chain-wide constant, which matters because the same item clears the bar in produce and fails it in center store. And because the post-reset measurement uses the same view, lost-item recovery at eight weeks is a read rather than a project, which is what makes the next range review better calibrated than the last.

Scout analyzes the range and models the cut. It is not a space-planning or planogram system: it does not draw the shelf or publish the reset to stores.

Summary

• The bottom three deciles are 30% of SKUs and 3.6% of dollars, but roughly half of them survive the four protective filters: destination items, under-distributed items, new items, and price-tier anchors.
• Cut where marginal contribution falls below the carrying cost of the space, which in center store at $288 per square foot lands around the eighth decile, deeper than most reviews go.
• Optimize additions too. Line extensions forecast as incremental and deliver as cannibalization, which is how a 240-item set becomes 310 with no more sales.

Further reading: shelf space optimization covers how much space each survivor should get, and sales per square foot is where the carrying-cost threshold comes from.