2
$\begingroup$

I have gotten a student job in the management department of a chain of 50 grocery stores. The job includes gathering daily statistics on the economy of the stores.

Every day a per-store revenue statistic is made, comparing the revenue to matching day last year (sunday/sunday, monday/monday) and an accumulated statistic for the month.

Also a per-store gross margin percentage statistic is made.

Now these are simple measures and I feel that there could be a lot more to be gained from the data, the data is specific down to the (cost price)/(sale price)/(number of sales)-level on every type of item.

Furthermore no or very few graphics are made and the data from the previous years are not taken into use.

Have any of you seen a similar problem? Do you have any ideas on how to proceed? Any easy-to-read, informative types of graphs you would want to share regarding these types of data?

$\endgroup$
  • 4
    $\begingroup$ It sounds as if your question is "What might I do with all this information?" Which is really not well suited to this Q&A site. There is no correct answer. Your best bet is to explore topics in statistics, especially in time series analysis. And you can do a fair amount of that by reading threads on this site. If you settle on a course of analysis and find that you're getting stuck somewhere, this will then be a good place to ask a question about it. $\endgroup$ – rolando2 Dec 11 '11 at 13:09
8
$\begingroup$

My present job involves analyzing a lot of retail data, so yes, I share your pain. To be clear, an issue for me is that there are extremely rich data but there is not much of an appetite for studying and presenting that richness. Here are some ideas to get started, but I'll start with a key rule I try to follow:

Always tie your analysis back to specific business decisions that will be influenced by what is learned

Even better, start with the business question of interest and let that guide your work. Otherwise, you risk being seen as the geeky data guy out of touch with the needs of the business.

Here are a few ideas that are focused on summarizing the chain. Presumably these summaries will lead to follow up questions, and that is where you can work with your colleagues to distinguish between analyses resulting in information that would be nice to know, versus analyses that affect specific decisions. You want to focus on the latter as much as possible.

  1. Find a nice way to summarize the per-store-revenue and margin statistics you have. For example, report the average, median, and some basic distributional information like min, max, and 25th/75th percentiles. You could create a simple "weekly scorecard" showing this information.

  2. Identify good/bad outlier stores. Is there anything in common within these subgroups (e.g. real estate quality, demographics, square footage)?

  3. Aggregate data at a monthly level and create a rolling 24 month time series for each store, as well as an overall series for the chain. Are there are any obvious patterns or trend?

  4. Depending on the geographical distribution of the 50 stores, it may be helpful to visualize where they are using Google Earth, say. You'll have to get lat/lon coordinates for each store, and you should also do this for key competitor stores. Then, visualize where they are and vary the size/color of the store symbols to convey profitability. Is there any pattern relative to region or competition?

Hope this helps.

$\endgroup$
  • $\begingroup$ (+1) very good ideas, but I especially like the point that data analysis should be connected to a (may be roughly) predetermined goal. $\endgroup$ – steffen Dec 13 '11 at 7:34
2
$\begingroup$

First, follow Josh's advice - starting with some summary statistics for the chain and for each store is the best way to let the data guide your inquiry.

That said, if the management of your grocery chain produces a planogram (Wikipedia, some sample images) dictating the physical placement of every product, you may want to try producing some visualizations incorporating that information. Making heatmaps of, for example, units sold and profit per product could help guide stocking decisions and product placement. Do the store's generic brands sell better above or beside their name-brand counterparts? Did last month's decision to move product X help or hurt sales of the goods across the aisle? Are there any unusual sales patterns that jump out?

Asking questions about product placement within stores may be a useful starting point because it's immediately actionable and low-risk: using your infographics to change the planogram for a few stores is easily reversed and incurs little additional cost. If your ideas result in increased sales, they can be deployed more widely.

$\endgroup$
  • 1
    $\begingroup$ 1)Do the store's generic brands sell better above or beside their name-brand counterparts? ... This is descriptive/arithemtic and lacks rigor 2)Did last month's decision to move product X help or hurt sales of the goods across the aisle?.... This is inferential requiring a model 3)Are there any unusual sales patterns that jump out? . This is inferential and requires a model $\endgroup$ – IrishStat Dec 11 '11 at 21:25
1
$\begingroup$

Your focus should be on detecting unusual and therefore by indirection the usual. Accountants summarize data and provide statistical descriptive analysis thus essentially "believing the data". Statisticians develop inferential tools that speak to challenging the data for consistency via hypothesis testing and predictive models. The models provide for the detection of consistency ( the signal/prediction ) and potential anomalies which may be useful to help discriminate between stores and/or line of business. These anomalies should be used to help carefully identify omitted variables/causes. The idea of looking for patterns for this is not new but rather underlies scientific reasoning. Performing aggregation or smoothing using any assumed model is at best descriptive. Objective statistical analysis ( meaning time series analysis in this case ) should be used to determine how to weigh the past for inferential purposes. It is well known that an ARIMA model is a super-set of simple assumed moving averages as it empirically finds the optimal number of weights to use and the optimal values of these weights rather than assuming (24) and a set of weights that are uniform (1/24). Allow the data to speak ! From http://books.google.com/books?id=SD_QX7x366MC&pg=PA6&lpg=PA6&dq=the+idea+was+presaged+long+ago+by+Francis+Bacon&source=bl&ots=43qzI3BOPo&sig=BKl9CEM7zLOUNv5s03Fynctss5Q&hl=en&ei=JQXmToChIpSutweKpumEBQ&sa=X&oi=book_result&ct=result&resnum=1&ved=0CB4Q6AEwAA#v=onepage&q=the%20idea%20was%20presaged%20long%20ago%

not my words but my mantra

$\endgroup$
  • $\begingroup$ What is the citation for your quote? Was it this book? $\endgroup$ – Andy W Dec 12 '11 at 13:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.