6. Finally
Zooming back out, and hiding all the rarely-bought items and weak connections
gives us an overview of the central, highly complex island. We see some of the
patterns occurring elsewhere - linked discounted hardware items and plumbing
items. We see expensive linked pieces of lumber (the solid square nodes
in the display) and we also see a very expensive piece of lawn/garden equipment
just top-right of center.
We have just started exploring this data set. We have a set of qualitative
results and questions for further analysis. There are many paths we could take
to delve deeper into this example. Here are a few questions we could try
answering:
- Who is buying discounted hardware items together? Is it contractors
(few items, but large numbers of them bought) or the general public
(maybe more items, usually only buys one of each type of hammer!)
- Why do plumbing and electrical departments have strongly different
discount and buying patterns? It seems they should be selling to the same
type of customer, so why are the patterns so dissimilar?
- What sort of items do people who buy expensive lawn tractors also
buy? Can we use those items to entice them to buy tractors?
- Noting the inter-departmental connections, does the store layout
reflect these patterns?
- Discounting which items would lead to the greatest increase in
sales for items bought on the same receipts as these items? Is there a
consistent and actionable pattern among them?
Nicheworks can be used by itself, or, as was done in this analysis, as
a component of the EDV system. Nicheworks is not limited in the number
of nodes and links it can handle, but practically it works best with
20 - 100 thousand of each, although it is still usable on a million.
gwills@research.bell-labs.com