Applying Science to Assortments

Understanding the performance of a retail assortment is more complicated than merely summing the sale of its component SKUs. Wallets are finite, and when shoppers choose one item to buy they may also be deciding not to purchase an alternative item. That is why it is an emerging best practice among leading merchants to apply customer-centric measures of transferable and incremental demand in the assortment decision.

Three branches of science
As a core concept, each item's incrementality is based on the number of "like" choices available to the shopper. Where there are many like alternatives, incrementality tends to be less; where there are few like alternatives, incrementality tends to be greater.

To apply these concepts to deliver the best possible local assortment, DemandTec applies three branches of science within an Assortment Optimization analytical model:

1. CONSUMER DECISION TREES
Consumer decision trees are a widely used method that examines loyalty and/or panel data to understand how shoppers choose between items at the category, segment and sub-segment levels. Decision trees also help retailers understand which products are in a shopper's consideration set when he or she comes into the store. Product groupings may be determined using statistical cluster analysis. Common attributes may be identified as a basis of customer segmentation. Many retailers already use decision trees routinely in their category planning activities. The Assortment Optimization analytical model can help to confirm or assess those as the process is initiated.

2. INCREMENTALITY CURVES
The second science, Incrementality Curves, creates representations of incrementality for each SKU. Combining these curves into a model allows retailers to study how consumers responded to past changes in the SKU count -- and to predict how they will react to changes in SKU count going forward.

In this diagram, the curve at the bottom shows a delisted SKU, selling 100 units per week until week 12. After delisting and sell-down of stock on hand, sales drop to zero for that item.

But the upper curve reveals that the total segment loses just 60 units per week, due to some switching or transference of demand to other items. (This analysis would work also in the opposite direction as SKUs are added.)

Studying these curves at the shopper segment level lets the retailer avoid removing items that are critical for core shopper segments.

3. OPTIMIZATION
The third science element DemandTec applies is Optimization. Put very simply, this consists of determining a desired outcome and letting the science built into the analytical model develop recommended steps to reach it.

Optimization goals may be set for incremental sales, profit, SKU counts, linear feet or other strategic objectives. Changing the parameters within the Assortment Optimization analytical model allows retailers to pose almost unlimited "what if" questions and predict results:

	What assortment will appeal best to the "budget family" shopper segment?
	What assortment will drive maximum volume in my stores?
	I want to reduce the amount of shelf space allocated to this category. What is the right assortment for the smaller set?

The Assortment Optimization analytical model gives retailers the ability to quickly determine better assortment solutions. It works efficiently at the chainwide level, but it's fast and flexible enough to support local variations by cluster, or even by store.