Identifying KVIs as a retailer is complex and hard, but it’s important enough that it must be done right. In the age of increased price comparison, online shopping, and showrooming, nailing the KVIs are how traditional retailers can compete on price while still maintaining margins.
We recently encountered another approach to KVI identification in the wild that used a broad set data sources, was analytically rigorous, and had gotten executive buy-in. Unfortunately, it was also dangerously biased and likely to yield misleading results.
The approach started well, relying on the usual suspects: basket metrics, and gap to competition. However, the customer association analysis they used was seriously flawed. And as is all too common in analytics, the entire approach overemphasized the analytic technique and lost sight of the business goal and implicit assumptions. To quote one of our favorite analytics professors (Hi, Prof. Powell!), the analysts had “done what their SAS package could do, rather than what they should do.” Here’s how it worked:
- They started with a good list of “Seed KVIs” that were reasonable and well-accepted.
- They then looked at which customers bought those “Seed KVIs”, and captured all other products those customers purchased.
- They ranked that list of products as candidates for KVIs across the store and rolled out aggressive new pricing.
The flaw is that they confused a KVI (an item which drives price perception) with a Price Sensitive Item (an item bought by price-conscious customers). To illustrate, imagine a supermarket near a college campus. Using the above approach, cheap ramen noodle packages, popular among cash-poor students, would float to the top of this list as a likely KVI. However, this universally cheap (and equally dubious) product is unlikely to have any influence on price perception. One of the necessary aspects of a KVI is high elasticity; reducing price needs to increase units sold by a lot. Since it’s also a good candidate to be a Giffen Good (all college students agree that it’s inferior!), its elasticity is likely low.
This is similar to what was found at the specialty retailer: whenever there was a “good-better-best” assortment, the algorithm routinely tagged the low-end of the assortment up as KVIs. This not only resulted in lost margin by reducing price on inelastic products, but the retailer also missed valuable opportunities to address the real KVIs in their assortments.
Postscript: Sigh. We later learned that the approach detailed above came directly from McKinsey, as detailed in their white paper here.