Assortment Reduction

What it is

A supermarket category can hold dozens of items that differ very little from one another. Peter Boatwright and Joseph Nunes opened their 2001 paper on that observation: most categories, they wrote, are cluttered with stock-keeping units, the individual items a retailer tracks, that differ little at the level of their attributes. A cut that removes that clutter looks free.

Retailers hesitated for a reason Susan Broniarczyk and her co-authors set out in 1998. Grocers had been told to reduce the number of items they carried, in line with consumer demand, under the label of efficient assortment. They resisted because they feared that removing items would lower how shoppers rated the assortment and make them less likely to choose the store.

There are two ways to describe a cut. One counts items and asks whether the ones removed were the low sellers or the favourites. The other counts attributes, such as brands, sizes and flavours, and asks which of those the range still offers. Boatwright and Nunes took the second view. They found customers reacted differently to the loss of sizes, but welcomed the removal of redundant items, and that brand and flavour were the two attributes that mattered most to them.

In effect

The evidence arrives in three stages. In 1998 Broniarczyk, Hoyer and McAlister tested two cues shoppers use to judge a category, whether a favourite is still available and how much shelf space the category keeps. Their conclusion was that retailers might make substantive reductions without harming assortment perceptions or store choice, as long as only low-preference items were eliminated and category space was held constant. Boatwright and Nunes later summarised the earlier work as showing cuts of up to 54 per cent of low-selling items with no significant effect on perceived variety or sales.

In 2001 Boatwright and Nunes analysed an online grocer in which 94 per cent of categories had seen dramatic cuts, particularly to low sellers. Across the 42 categories they examined, sales rose by an average of 11 per cent. Sales rose in more than two thirds of those categories, and 75 per cent of households increased their overall spending. They also found that category sales depended on the total number of items offered, which ran against the earlier work.

In 2005 Sharad Borle, Boatwright, Joseph Kadane, Nunes and Galit Shmueli returned to household records from the same grocer, this time covering every category in the store. The cuts ran from 24 to 91 per cent of the items in a category. The 840 households offered the reduced range were expected to wait about 25 per cent longer between orders and to spend 4.8 per cent less per order than the 378 households that kept the full range, and the fall in order frequency cost more than the fall in order size. Categories bought less often lost the most. Staples such as milk, bread and cereal lost the least. The authors reported the 2001 study, together with a 2004 correction note, as finding no change in sales, an estimated rise of 1.5 per cent with a standard error of 1.4, and attributed the difference to the 2001 analysis using a subset of categories and households that leaned towards frequently bought staples. Fitted to the 2001 households but with all categories, their model returned the new result.

What it does not say

It does not say that cutting a range is a mistake. The 1998 work found conditions under which it was harmless, and the 2001 paper found that shoppers welcomed the loss of redundant items. The conditions are what matter: which items go, and what the shelf looks like afterwards.

It does not say the 11 per cent figure stands. The 2005 paper reports the 2001 study as showing no change, and the 2004 correction note could not be read for this page, so what it changed in the original figure is not established here.

It does not say the 2005 result applies to a whole market. The principal experiment was run in a single city in which delivery fees did not vary, and it measured changes in how existing customers behaved, not whether a thinner range would fail to attract new ones.

And none of the studies tested a menu or a product page. For those, the grocery results describe a model of what to look for, not a measurement. For the jam study that started the wider argument, and the meta-analyses that followed it, see the article on choice overload after the jam study.

---