The comparable set is a judgment call, and it moves. A store closed six weeks for a remodel either stays in the set and drags the number down, or leaves it and flatters it, and whichever you choose, prior periods rarely get restated to match. Every retailer picks its own qualifying period, so comps from two companies are not comparable no matter how confidently they are placed side by side. And once online orders are fulfilled from store stock, whether those sales sit inside comps changes the answer materially.

Mixing bases. Sales are measured at retail value, inventory is usually held at cost, and dividing one by the other gives a number wrong by your entire margin. The second trap is average inventory taken from just the opening and closing balance, which for anything seasonal misses the peak completely and overstates turns. Measured across the whole assortment it also conceals the usual shape of the problem, where the top decile turns fine and the tail has not moved in a year.

Reported at company level it is meaningless, because a high-margin slow category and a low-margin fast one average into a figure no decision can be made from. It is also sensitive to how inventory is allocated when the same item sits in more than one category, and to whether markdowns reduce margin or sit in a separate cost line. Two teams can compute GMROI honestly on the same data and disagree by a wide margin because of that one treatment choice.

Reported at the end of a season it always looks acceptable, because everything eventually sells at some price. The number that matters is sell-through at full price, before the first markdown, and that one is usually far worse. Measuring against units received rather than units actually available also flatters the figure, since stock still in transit or sitting in a distribution center never had a chance to sell.

You only learn it at physical count, so by the time it surfaces it covers months and cannot be attributed to a cause, a week or a person. It also combines four unrelated problems into one number: external theft, internal theft, damage, and administrative error. Each has a different fix, and averaging them is how a business ends up spending on security cameras when the real problem was in receiving.

The denominator comes from a door counter, and door counters count everything: staff arriving, deliveries, someone stepping outside to take a call and walking back in, a family of four counted as four shoppers with one basket between them. Because conversion is a small number, a small percentage error in traffic produces a large percentage error in conversion. Comparing stores with different entrance layouts or counter hardware compares the sensors rather than the stores.

It moves for two entirely different reasons, price and quantity, and cannot tell you which. A promotion that adds units while cutting price leaves ATV flat while margin falls, which reads on the report as nothing having happened. Read on its own it also quietly rewards discouraging small transactions, which is almost never the strategy anyone chose.

Multi-buy promotions inflate it mechanically without anyone having sold better, so a strong UPT month can be a pricing decision rather than a selling improvement. Returns processed as separate transactions distort both this and average transaction value, usually in opposite directions. Measured across formats or categories with very different basket shapes, it averages things that were never comparable.

The denominator has no standard definition. Selling floor only, or including stockroom, fitting rooms and back office? Two stores performing identically can report very different numbers. The larger distortion now is fulfillment. Orders bought online and collected or shipped from a store either credit that store's productivity or do not, and that choice alone can swing the figure far enough to change a decision about closing the location.

Most retailers substitute system-based out-of-stock reporting, which is blind to the most common failure. The system says twelve units on hand, and all twelve are in the back room, misplaced on the wrong shelf, or never arrived because of a receiving error. That phantom inventory is invisible to the report and completely visible to the customer standing in front of an empty facing. Availability measured at a distribution center says nothing at all about the shelf.

It is quoted two ways, as a share of sales and as a share of original retail value, and the same underlying data produces very different figures depending on which. Promotional markdown, clearance markdown and permanent price reductions get added together even though they signal completely different problems, one of them deliberate and two of them not. And reported after the season it is a post-mortem; the version that could have helped is the markdown still to come on stock you are holding right now.

The rate of sale is almost always a trailing average, which creates a circular trap: a fast item that just sold out shows a low rate of sale precisely because there was nothing left to buy, so it reads as adequately stocked at the exact moment it is starved. The same flat average also erases seasonality, so a Christmas line looks catastrophically overstocked in October and perfectly fine in mid-December when it is far too late to reorder.