A nice article from the Washington Post (incidentally, owned by Jeff Bezos) about Whole Foods (now part of Amazon). When the acquisition was announced, the press was feeding us stories that Amazon intends to drop prices at Whole Foods in a turnaround bid. It was a heart-warming story but is it true?
The research firm Gordon Haskett investigated. Their conclusion: Prices were on average down only 1.2%. That is a much lower percentage drop than the press was (mis)leading us to believe. The article states that Whole Foods prices were typically 15 percent higher (pre-acquisition).
This top-line change rate is what appears in news headlines. But the devil is in the details. What kind of average is it?
First, it has to be an average across items in the store. But this isn't an average of all items on sale - that would be an almost impossible data collection exercise for an outsider to complete. According to GH, this is an average of about 115 items. (I have requested a copy of the report as it sits behind a wall.) This is actually a tiny proportion of the total items sold. According to this industry association, a supermarket typically stocks almost 40,000 items.
Second, it should be an average across all stores in the U.S., perhaps weighted by their sales. Alas, this is not the case here. GH said they took data from a single store, in Princeton, NJ.
Third, it is an average across weeks. The prices of the 115 items were checked each week for five weeks. One of the key findings is that prices have been creeping back up.
The most important paragraph in the Washington Post article is the last one, which includes a quote from David Livingston, a research analyst focusing on supermarkets. "Like all grocery stores, if prices are lowered someplace, prices are raised someplace else. The overall change is nil."
Now, this statement implies that the average change should be 0% but the study returned the number 1.2% - why the difference? It's hard to say for sure given that I haven't received the research report yet. But I can speculate.
Think about which 115 products should be on the pricing survey. There are two popular ways to sample these items from the population of 40,000 items. One is random sampling; the other is a targeted sample - perhaps focusing on the most frequently purchased items. It's quite possible that the latter sampling method was utilized. Another industry expert spoke about the following: "The whole game is that you want the 100 most recognizable things — milk, apples, bananas — to be cheaper. If you can do that, you can build a perception that the whole store is competitively priced." Given this type of thinking, it might make sense to do a targeted sample.
Neither of these methods is perfect. Random sampling off a product list almost guarantees that there will be a good number of sampled items that sell few units per week. Those items would not be representative of the shopper's experience if few shoppers buy them. Targeted sampling may yield a more useful answer. Of course, there is a BUT. Read the above pricing strategy again - it sounds like a sample of 100 most recognizable things is a biased sample - these products are competitively priced and frequently discounted (the so-called "loss leader" strategy to bring people into the stores.) If these prices already have razor-thin margins, there is little room for Amazon Whole Foods to lower these prices so we expect the average price change for these items to be smaller than the overall average price change.
So, can we believe those GH numbers? Only if (a) we think those 115 items represent a useful, and valid sample and (b) we believe that Princeton, NJ is the "average" store [ or, that Whole Foods does not vary their price change strategy by location. ]