Using Association Rules for Product Assortment Decisions in Automated Convenience Stores

1
Using Association Rules for Product Assortment
Decisions in Automated Convenience Stores

• Weizheng Gao
• 2003/3/20

2
Outline

• The fully-automated convenience store
• PROFSET a product selection model
• Empirical study
• Conclusions
• Questions and Comments

3
?. The fully-automated convenience store

• What is fully-automated convenience (FAC) store?
• Specific problems in terms of assortment analysis
  and product placement.

The product assortment is limited to a
selection of about 200 products. Customers
typically buy only one to three different
products during a single shopping occasion.
4
?. PROFSET a product selection model

• Problem situation

Optimal product assortment should meet two
important criteria
1. The assortment should be qualitatively
consistent with the stores image.
basic products and added products 2. The
assortment should be quantitatively appealing in
terms of the profitability it generates for the
retailer.
5
?. PROFSET a product selection model (cont)

• PROFSET model for product selection based on
  frequent sets

? product-specific profitability heuristic
Model parameters Model specification
6
Model parameters

• Gross margin

Let Tj an individual sales transaction at time
j SPi be the selling price of product i
PPi the purchase price of product i
fi the number of times product i was purchased in
Tj
? mTj is the gross margin generated by sales
transaction Tj
7
Model parameters
Model parameters

• Gross margin

• ? Mx is the gross margin of frequent itemset X

• Costs (of products)

Product costs should be included.
A total cost figure Ci per product i
8
Model specification
9
?. Empirical study

• The empirical study is based on a data set of
  27,148 sales transactions over 5.5 months. The
  product assortment of the store consists of 206
  different items. The average sales transaction
  contains 1.4 different items. And costs are not
  included in the model.

Two important phases
1. Purchase behavior under the form of frequent
itemsets is discovered by using association
rules. 2. The PROFSET method is used to select a
hitlist of products from the assortment.
10
?. Empirical study

• Mining for association rules

An absolute support of 10 was chosen.
523 frequent itemsets were obtained of size 1
or 2 with absolute support ranging from 10 to
2833.
11
?. Empirical study

• Product selection (PROFSET)

Experiments with the PROFSET method, result can
be identified
1. Some products with low product-specific
profitability but high cross-selling effects are
selected.
2. The PROFSET method allows to identify the
importance of the impact of such decisions on the
total profitability of the hitlist.
12
?. Empirical study

• Observation 1

Table 1 total margin, position and selection for
brand x and y
Cigarette paper ? tobacco support 291,
confidence 1.00
Tobacco ? cigarette paper support 291,
confidence 0.82
13
?. Empirical study
Definition
Table 2 economic interpretation of interest
tobacco and cigarette paper interest is 76.7 gtgt 1
14
?. Empirical study

• Observation 2

The impact on total profitability caused by
product-mix decisions can easily be assessed by
means of sensitivity analysis.
15
?. Conclusions

• The purchase data validate this approach against
  a frequently used heuristic for product
  selection.
• Results indicated that this model can identify
  cross-selling effects by using frequent itemsets.
• Analysis could easily be carried out enabling the
  retailer to assess the profitability impact of
  product assortment decisions.

16
Questions and Comments

• Thank you!