In the big town near my home town there are two supermarkets : Woolworths (W) and Coles (C) – (actually BiLo, a sub-brand).

Casual observation suggests very different levels of sales – W seems much busier than C.

Of course this observation may be incorrect, as we tend to shop on certain days at certain times, but let us treat it as if it is the truth. If we are not prepared to do this, a decent measurement of the magnitude of the difference could be readily ascertained by an appropriate set of observations using a structured design to even out any time related variability.

Now why this difference in sales should be so, and what I can do about it, would be questions of great interest to me if I were the capo di capi at C (or at W) : even more so if the pattern was replicated elsewhere.

What sort of data would I need, and how would I go about analyzing it?

Let’s not immediately leap for the consumer survey tool.

It may be that we need to input consumer perceptions into the model, but there is a lot we can do with just some careful observation and sensible analyses.

We can think of a whole bunch of reasons why this phenomenon might be occurring .. let’s put down a few, then see whether we can classify them into those that we can reasonably observe or assign measurements to ourselves and those for which we have to ask shoppers.

There is not a great deal wrong with deferring the more complex survey option until later .. we can consider our analysis as 2 stage, with the first stage being done on the basis of our ”expert” (as in sensible, straightforward) derived data .. with the residuals, the unexplained variation, being addressed in a later stage (if needed).

Some Hypotheses and a Draft Concourse

OK, some hypotheses:

• parking is marginally inferior at C
• range of goods is about the same in each
• prices .. comparable, but needs confirmation
• store size and spaciousness .. C marginally better, bigger aisles
• atmosphere – generally louder and busier at W
• lighting .. duller at C
• popularity .. most of my social circle shops at ..
• socioeconomic perceptions .. C is arguably downmarket
• checkout machines .. better at C
• checkout staff .. better at C
• nearby shops . a few more at W
• history .. C established later than W
• origins .. C is perhaps perceived as “foreign” (from another state)
• liquor .. W has a liquor store attached although it does not seem to do much business, and there are other convenient liquor stores in town

Well, we could go on. To get a better and more systematic “concourse” (set of evaluation criteria) we can certainly review some past work, and I personally have done choice modelling work in the supermarket choice area.

However we have to be careful to ensure that we do not get too abstract and we should at least be open to the possibility that store choice is driven by local and idiosyncratic factors.

Collecting measurements of C and W on the above factors should not be too hard. We can use some mystery shoppers, or trained observers to do this. And, of course we want some data on the relative sales of C and W, but a reasonable proxy for that would be a careful estimate of the number of shoppers. And we want to replicate this over locations where we do have similar competitive environments i.e. both C and W trading : effectively we want to do twin studies.

Analysis .. well there is a bit of normalization to be done (we don’t care about absolute sales, merely the ratio of sales or somesuch) , and thence we simply fit a linear model.

Simple enough, likely to be informative and actionable, and relatively low cost.

Grand Strategies, and the “I think ..” factor : the antithesis of Data Analytics

As I write this, Wesframers has launched a takeover bid for Coles and this has of course brought the pundits out in droves. It is proposed that Coles needs more “customer-focused management”, needs to be more of a “fresh food specialist”, needs to change its “corporate culture”, needs to “release its businesses from centralized corporate structures”, “improve value and freshness and availability”, need to “feel part of the neigborhood”, “make the shopping experience more enjoyable”, “leapfrog Woolworths in the delivery of fresh food and to target European-style quality and reverence for the product”, “bring food to consumers at the best possible price” etc etc

Since there is little or no evidence (as in real data) in support of the likely efficacy of otherwise of any of these nostrums, no basis for believing that any of these are real expressions of the true problems and manifest difficulties in turning any of this overblown pundit-speak into anything that could be measured let alone input as explanators into some believable model, then one is left feeling distinctly uncomfortable.

A litle bit of KISS would help : start with some simple data that is operationizable and actionable, measurable and modellable. From that first simple model, by an examination of where it goes wrong and fails to predict the obseved differences correctly, form some additional hypotheses and extend the model. Repeat.

Simple, no?