We offer a framework for conducting market basket analysis (MBA), a powerful technique for determining retail association rules and other patterns that can help retailers increase profits and customer loyalty.
The Work Ahead in Intelligent Automation: Coping with Complexity in a Post-Pa...
Unraveling the Customer Mind
1. • Cognizant 20-20 Insights
Unraveling the Customer Mind
A soft economy and unrelenting margin pressures require retailers to
think more proactively about what drives consumer buying preferences
and behaviors. Our unique twist on market basket analysis can help
retailers create more personalized campaigns that result in greater
transactional activity and deeper customer loyalty.
Executive Summary highly variable. Thus it is important to understand
how budgets are being used, what impact it has
As the economy slowly recovers and a raft of
on your organization/business and how to ensure
new channels, products and offers emerge, some
success within this changing landscape. Tradition-
consumer segments (particularly the affluent)
ally, retailers spend a significant amount of their
are gradually spending more money at retail
campaign budgets on TV, followed by print, online,
stores. It is therefore imperative for retailers to
catalogues, direct marketing, events and outdoor
better understand consumer thinking and derive
advertising. However, the pattern is shifting
behavioral insights that reveal buying prefer-
toward greater investments in online channels.
ences and decisions that take advantage of this
The retail industry is expected to be one of the
rejuvenated willingness to consume.
biggest spenders on online advertising, according
Consumers are ever-changing. In fact, research to researchers who follow the space.
(both hard data and anecdotal evidence) shows
The industry is also highly competitive and
they are better informed, more nimble, more
fragmented globally, so it is increasingly
selective and less loyal than ever before. To win
important to capture and analyze every customer
the hearts and minds of consumers, billions of
interaction. Moreover, with the improved sophisti-
dollars are spent annually on mass advertis-
cation of data handling and collection technology,
ing and promotional campaigns both online and
retailers have access to a gold mine of POS
offline. These campaigns are not targeted to
consumer transactional data across multiple
specific customer segments nor are they linked to
channels.
customers’ online and offline purchasing behavior
because most retailers lack an in-depth under- The struggle for retailers is to differentiate
standing of consumer preferences and purchasing themselves through their product offerings and
behavior. This results in campaigns that bombard promotions to customers across all channels.
consumers with offers, discounts and promotions Some smart retailers effectively deploy targeted
at odds with their interests and needs. product offerings, which can generate significant
revenues; poorly chosen ones, though, waste
Within this changing landscape, both marketing
money and opportunity. For example, grocery
strategies/options and marketing budgets are
cognizant 20-20 insights | december 2012
2. chain operator Kroger recently refined its direct long did it take to tend? How many items — by
marketing strategy by using data from its loyalty- type — were in the basket? What was the relation-
card program and sent unique coupon offerings to ship among the purchased items? How do other
specific households. In the words of David Dillon, baskets compare?
Kroger’s chairman-CEO, “We
The reasons why most retailers refrain from a
The reasons why understand and appreciate that deep-dive analysis is their preconceived notion
most retailers refrain no two customers are alike.” The that analyzing data at this level of granularity is
company believes that this level
from a deep-dive of promotional personalization expensive, overly time-consuming and has limited
business value.
analysis is their offers a path toward creating a
preconceived notion direct link to its customers that Solutions Emerge
no other U.S. grocery retailer
that analyzing can replicate.1 Effective retailing requires an immediate response
data at this level to consumer requirements, mandating extremely
This white paper lays out new efficient, agile and responsive business and oper-
of granularity is thinking about market basket ational processes. Store operations, merchandis-
expensive, overly analysis that can help retailers ing, marketing and advertising must all perform
time-consuming drive campaign effectiveness in consistently, with little room for error. Due to con-
more tailored ways and generate tinually increasing pressure on margins in many
and has limited greater per-customer purchases industry segments, only the best retailers are
business value. and loyalty. surviving; few are thriving.
In Search of Enlightenment To execute on all cylinders, retailers must gain an
in-depth understanding of their operations and
Since the introduction of electronic point of sale,
maintain the ability to delve into the operational
retailers have had at their disposal an incredible
data to ask (and answer) any and all business-
amount of data. The sheer volume of data,
critical questions. Traditional DW/BI systems were
however, obscures patterns, making it impossible
designed to handle large amounts of summarized
to discern customer preferences and behavior via
information to address a predefined set of
manual inspection. The initial challenge, tradi-
questions. To be agile and adaptable, retailers
tionally, has been to seek ways to leverage trans-
need the ability to continually measure, track and
actional data to produce business value. Most
probe all aspects of their businesses to answer
retailers have already figured out a way to con-
new questions.
solidate and aggregate their data to understand
the basics of the business — what is selling, how This starts with effective market basket analysis.
many units are moving and the sales amount. Figure 1 depicts a shopping cart by a typical
However, few retailers are successfully analyzing consumer containing various purchased products.
this data at its lowest level of granularity: the A complete list of purchases made by all customers
market basket transaction. provides much more information; it describes the
most important part of a retailing business — what
In fact, the essence of the analysis lies in the
merchandise customers are buying and when.
depths of the detail. Thus, a 360-degree view
The analysis uses the information about what
of the customer through integrated online and
customers purchase to provide insight into who
offline transactional data can ensure retailers
they are and why they make certain purchases.
that the products they offer and promotions they
run match shopper preferences and behavior Market basket analysis provides insight into the
and deliver maximum return on their marketing merchandise by highlighting which products
spend. The ability to link purchases to individual tend to be purchased together and which are
purchasers can take this even further; for most amenable to promotion. This information is
example, tailoring offers to specific customer actionable; it can be used to:
segments and driving higher returns from more
precisely targeted campaigns.
• Understand new store layouts.
• Determine which products to put on special
For retailing, this would mean understanding promotions and bundles.
everything possible about the sales transaction,
including: What time of day did the customer
• Indicate when to issue coupons.
shop? How long did it take to check out? Was a The data-mining technique most closely allied
loyalty card used? Who was the cashier? How with this analytical approach to market basket
cognizant 20-20 insights 2
3. Market Basket Analysis Unravels Customer Purchasing Behavior
In this shopping basket, the shopper placed a quart
of orange juice, some bananas, dish detergent, some
window cleaner and a six-pack of soda.
Is soda typically purchased with
Are window cleaning products bananas? Does the brand of
purchased when detergent soda make a difference?
and orange juice are bought
together?
How do the demographics of What should be in the
the neighborhood affect basket but is not?
what customers buy?
Figure 1
analysis is the automatic generation of associa- • Choosing the right set of items.
tion rules. Association rules represent patterns in
the data without a specified target. As such, they
• Generating rules by deciphering the counts in
the co-occurrence matrix.
are an example of undirected data mining.2
• Overcoming the practical limits imposed by
In a relational database, the data structure thousands or tens of thousands of items.
for market basket data has the following key
components: A word of caution regarding association rules:
Retailers need clarity and utility of the results,
• The order is the fundamental data structure for which are in the form of rules about groups of
market basket data. products. There is an intuitive appeal to an asso-
ciation rule because it expresses how tangible
• Individual items in the order are represented products and services group together.
separately as line items.
• Product reference tables provide more descrip-
tive information about each product. This A Co-occurrence Three-
should include the product hierarchy and other dimensional Matrix
information that might prove valuable for
analysis.
• The customer table is an optional table and
should be available when a customer can be
identified.
Detergent 1 0 0 1 1
Association rules were originally derived from
Soda 2 0 0 2 1
point-of-sale data that describes what products
are purchased together. The sheer bulk of this Milk 1 1 1 0 0
transactional data (see Figure 2) — recording, at
the item level, every purchase through stores, Cleaner 1 1 1 0 0
online storefronts and other channels – makes it Detergent
Soda
very hard to understand the repeated patterns of OJ 4 1 1 2 1 Milk
Cleaner
purchasing that provide insights into customer OJ
OJ Cleaner Milk Soda Detergent
behavior and preferences.
Orange juice, milk and window cleaner
The basic process for finding association rules is appear together in exactly one transaction.
illustrated in Figure 3. There are three important
steps for creating such rules: Figure 2
cognizant 20-20 insights 3
4. Three Basic Steps for Building Association Rules
1 First, determine the right set of items
and the right level. For instance, is pizza
an item or are the toppings items?
Topping Probability
2
Next, calculate the probabilities and joint probabilities
of items and combinations of interest, perhaps limiting
the search by using thresholds on support or value.
If mushroom, then pepperoni!
3 Finally, analyze the probabilities
to determine the right rules.
Figure 3
We created a market basket analysis (MBA) • Increasing the size and value of the market
solution framework to help clients demystify this basket.
data and provide actionable insights into customer
purchasing behavior and their responsiveness
• Testing and learning by using the marketplace
as a laboratory.
to certain channels/promotions. Although their
roots are in analyzing point-of-sale transactions, • Determining the “magic” price points for this
store.
association rules can be applied outside the retail
industry to find relationships among other types • Matching inventory to need by customizing
of “baskets.” Examples of potential cross-industry store and assortment to trade area demo-
applications include: graphics.
• Items purchased on a credit card, such as rental • Optimizing store layout.
cars and hotel rooms. Thus, pattern analysis can be used to drive
• Information on value-added services pur- decisions on how to differentiate store assortment
chased by telecom customers (call waiting, call and merchandise as well as to effectively combine
forwarding, DSL, speed call, etc.) can help oper- offers of multiple products, within and across
ators determine how to improve their bundling categories, to drive higher sales and profits.
of service packages. These decisions can be implemented across an
• Unusual combinations of insurance claims can entire retail chain, by channel or, if the data is
be a sign of fraud. analyzed at the store level, customized offers can
be formulated and deployed at a local level.
Retailers’ Use of MBA Analysis
It is imperative to understand the critical missing
Retailers use MBA analysis to explore transaction
link; traditional basket analysis fails to provide
data to determine the affinities of what people
actionable insights. However, when done correctly,
buy to detect changes in basket composition,
MBA can uncover how effective promotion of
size and value, and to discover new insights into
a given product can increase sales and profit
customer buying behavior. This includes:
influenced by related products and drive larger
• Identifying more profitable advertising and baskets or greater transaction volumes. Hence,
promotions. our approach is aimed at revealing transaction
• Targeting offers more precisely to improve ROI. patterns, illuminating key cause-and-effect rela-
tionships, to help retailers isolate the incremen-
• Generating better loyalty card promotions with tal impact of any given promotion (i.e., transac-
longitudinal analysis.
tions that wouldn’t have occurred but for the
• Attracting more traffic into the store. promotion). Affinity analysis3 on its own cannot
cognizant 20-20 insights 4
5. accomplish this; however, conducting a test vs. milk at an aggressive every-day low price (EDLP);
control measurement along with market basket sell-through is high and sales look good. And the
analysis helps to unravel the cause-and-effect company is confident that the sacrificed margin is
relationships related to the incremental impacts justified as it must be driving traffic to its stores
of promotions. and generating incremen-
tal sales of other items. Predictive models helps
The analysis framework links actions to outcomes However, upon looking more retailers to direct the
and empowers the retailers to truly understand closely at the baskets that
the mind of the consumer. The solution contain milk, the retailer
right offer to the right
framework is tool agnostic and can be deployed realizes that those baskets customer segments/
as an analytical layer to a vast majority of COTS tend to be single-SKU, or profiles, as well as
tools available in the market. Personalization and otherwise small baskets. In
tailor-made campaigns are all the rage in retail reality, the pricing strategy
gain understanding on
campaigns and promotions. In fact, gone are the was not at all efficient, what is valid for which
days when one-size-fits-all large promotions are but this would have been customer, predict the
applied to increase basket size. With an increase impossible to determine
in margin pressures, marketers are trying to focus without gaining visibility
probability score of
on getting that extra mile. into the market basket. customers responding
MBA analysis will help retailers refine their Using this insight, the to that offer and
retailer decides to raise its
approach to drive an effective loyalty card scheme
EDLP on milk. The retailer
understand the
or online shopping registration by bringing a
personal touch. Combining POS data with other expects sales of milk to customer value gain
geographical level demographic information, drop, and it may even lose from offer acceptance.
customer interaction information across channels some customers. But those
such as e-commerce, loyalty club Web sites or customers were not profitable; the improved
order or service hotlines, as well as attitudinal margin on the future milk sales will result in
data captured through surveys at points of inter- profits being net-positive.
action, is sifted and analyzed to provide valuable
MBA empowers merchants to buy smarter
insights to further refine targeting strategy.
and strengthen their negotiating position with
Further, predictive models are built on historical vendors by providing the merchants with better
purchase data, as well as other attribute data, to information about customer buying behavior.
add predictability to customer responsiveness to
Integrated Solution Framework for Effective
promotions. This empowers retailers to embrace
Decision Making
focused targeting. Predictive models help retailers
to direct the right offer to the right customer Our MBA solution offering is coupled with an
segments/profiles, as well as gain understanding integrated analytics and BI platform called
on what is valid for which customer, predict the iTrackTM that integrates multiple stakeholder
probability score of customers responding to that metrics and views with a robust data model to
offer and understand the customer value gain generate business insights for effective decision
from offer acceptance. making. Together they provide:
MBA Helps Merchandisers • An integrated view of business metrics
across dimensions, with embedded roles and
Merchandisers need to see long-term trends to
privileges.
decide how much to buy and how the assortment
fits into the business model. Here are some • A single version of truth.
ways that leading retailers can leverage MBA to • An end-to-end platform and process from data
empower their merchandisers: acquisition to reporting.
• Can the retailer sell fast enough to cover car- • Multipledata sources integrated to create a
rying costs? comprehensive data mart for mining.
• Will the initial markup provide sufficient margins • Visibility to a comprehensive set of business
to promote sell-through? metrics.
These high-value, high-risk decisions can be sig-
• Informed decision making through insights and
report comparison.
nificantly improved with the customer insights
provided by MBA. For example, a retailer sells • Report customization.
cognizant 20-20 insights 5