Você está na página 1de 14

Tractor Supply Company

Analytical Proposal for 2013


October 2012

LETS BE DIRECT

Situation
SourceLink and Tractor Supply Company have a long history of being trusted partners. As such,
SourceLink is committed to helping Tractor Supply achieve and exceed its corporate goals and
initiatives. Tractor Supply has identified five objectives for corporate focus and improvement.
Two of these objectives are enhancing customer insights and optimizing earnings. With this
proposal, SourceLink will outline how we can help Tractor Supply leverage analytics to achieve
these objectives. SourceLink understands and wishes to facilitate the needs of both the TSC
Marketing and the TSC Finance teams. We have outlined several initiatives that we feel would
help TSC achieve their goals for 2013 the Sister Store Segmentation Study, Store Intercept
Attribution Study, a New Store Launch Smart Model, Campaign Optimizer Multivariate Testing,
Secondary Customer Profiling, and a custom built Customer Segmentation.

SourceLinks Customer Intelligence Tools


The success of analytics is predicated on data accuracy, competence in relevant
methodology and adherence to proven process and approach. SourceLink has a
distinguished track record of working with clients to develop superior direct marketing
strategies, programs and campaigns. Our strength is helping clients identify specific
opportunities within high-potential target markets that result in sales growth and
increased profits. We develop customer solutions based on the objectives of our clients,
rather than applying a one-size-fits-all approach. SourceLink offers innovative strategic
approaches and methodologies to all direct marketing channels, particularly direct mail.
We are passionate about the use of customer intelligence that informs business and
marketing decisions to realize Return on Marketing Investment.
Measuring, analyzing and reporting on all relevant program results contribute to the
refinement of subsequent marketing initiatives. SourceLink provides a breadth of
customer database analyses that form the basis of modeling for cross-sell, retention and
acquisition programs. SourceLink utilizes CRISP (Cross Industry Standard Process for
Data Mining). When combined with our proprietary techniques, CRISP ensures best in
class modeling and analytic results. The following chart outlines this rigorous
methodology. SourceLink utilizes CRISP (Cross Industry Standard Process for Data Mining).

LETS BE DIRECT

When combined with our proprietary techniques, CRISP ensures best in class modeling and
analytic results. The following chart outlines this rigorous methodology.

Phase

Activity Details

Business

Establish business and modeling objectives

Understanding

Familiarization with client business

Data
Understanding

Identify key data sources available


Identify in detail data fields that will be available and can be
used for building the model
Clean the data and format the data
Outlier and Missing Values Treatment

Data Preparation

Create modeling dataset


Create appropriate target variables based on demographic,
geographic, product or service and transaction related variables
Single View Foundation for model development
Exploratory Analysis, Hypothesis Building

Model Development

Univariate analysis
Develop model on development sample
Validate model on hold-out sample

SourceLinks National Consumer Database


Successful direct marketing programs start and end with the data. That's why SourceLink
continually invests in building and maintaining our National Consumer Database, giving
our clients unprecedented insight into their highest opportunity customer and prospect
audiences.

LETS BE DIRECT

SourceLinks National Consumer Database is our proprietary, enhanced database,


created by us and not sold or used by the wholesale data resource community. We
have more than 200 million consumer records populated on our database. SourceLink
doesnt take compiled data at face value. We analyze it to determine which elements are
most accurate and have the most coverage. This assists in overcoming flaws in
individual data sources and improves overall performance. SourceLink continuously
provides demographic enhancements by using a variety of data sources to crossvalidate information, fill in missing data and continuously model for the most
comprehensive coverage. By way of substantiating the value of our enhanced
demographics SourceLink has utilized validated client customer information for
comparison purposes to contrast typical raw data on national files versus SourceLinks
hybrid of multiple sources and models. Key variables such as Home Ownership,
Marital Status and Children in the Home are examples of demographics that
significantly improve modeling and segmentation results when SourceLink enhanced
data variables are used.
Data richness and accuracy are the strength and foundation of our analytics. This
includes our rebuilding and updating of the following variables nationally:

Home Value

Presence of Children in the Home

Household Income

Marital Status

Home Ownership

New for 2013, we will have enhancements to


o

Age (including exact date of birth for birthday campaigns)

Financial Health Value Scores

Automotive Data (Identifying truck owners)

Enhancements to Information on Children in the Household

Environmentally Conscious Lifestyle (secondary customer profile).

Presence of Smart Phone (text promotions)

Higher Population - Hunting/Shooting & Boating/Fishing/Sailing Lifestyles

LETS BE DIRECT

Enhancements to Household Income

Niche/SuperNiche Customer Segments rebuilt and enhanced (consider


testing against Mosaic Codes)

We are also sourcing a new B2B database built on Dun & Bradstreet data for better
business targeting. This could be leveraged for campaigns to identify or acquire tax
exempt agriculture customers.

I.

Sister Store Segmentation Schema

 Glossary of Terms:
To facilitate a shared language for the project, below is a glossary of terms.

o
o
o

Dimension A single variable or attribute.


Cluster A single cell representing a grouping of like stores.
Segmentation A group of clusters identified using K-Means clustering
methodology.

Segmentation Schema The entire segmentation approach, which could


include multiple segmentations.

Segmentation Matrix A matrix with each cell cross sections of the


segmentations being used for an initiative. All cells will not necessarily be used.

 Store Segmentation Schema:


SourceLink will work with TSC to determine the most appropriate TSC data to include in
the analysis. Data elements to consider include demographic data, competitive market
data, transaction details and store data including formats, age, performance, price
zones, and customer satisfaction data. Using standard clustering methodologies,
SourceLink will assign each store to a single cluster within each segmentation.
Segmentation is as much art as science. SourceLink anticipates identifying several
possible Segmentation Schemas that would be useful for specific needs such as geodemographic clusters, geo-competitive clusters, and product driven clusters. While this
seems complex, it is very actionable and flexible. For example, if a schema is
developed with 6 geo-demographic segments, 5 product segments, and 3 geo-

LETS BE DIRECT

competitive segments, the results would include 3, 5, 6, 15, 18, 30 or 90 cells in a


segmentation matrix for any application. All cells could be leveraged in the matrix or
efforts could be focused on certain cells.
Geo-Competitive Segmentation

Geo-Demographic Segmentation

Product/Transactional Segmentation

LETS BE DIRECT

 Sister Store Pairings:


The Store Segmentation Schema can be leveraged to select sister store pairings for test
and control treatments. SourceLink recommends that pairs be chosen within a single
cluster incorporating secondary factor(s) based on the intended test or use. As part of
the roll out, SourceLink will provide detailed profiles ranking each cluster top to bottom
by agreed upon key variables. This will further refine the process to choose sister store
pairings within the clusters. For example, if a cluster has 50 stores and TSC would like
to run a campaign in which household income is likely to be a key driver, a pair of stores
can be selected from the same cluster, which is most similar to one another on the
dimension of household income. This approach allows for maximum flexibility to change
pairings as objectives change.
 Potential Uses of Segmentation:
o

Promotions Testing DM, e-mail, publications, TV, radio, web, etc.

Promotions Targeting

Cross-Sell / Upsell Strategies and Targeting

Creative Marketing Design & Messaging

New Store Launch Strategy

Identify Challenged or Declining Stores

Improved Customer Service

 Segmentation Methodology & Analysis:


SourceLink MicroMarkets is a proprietary product utilizing a series of analytical
processes, over 1 Million data points, and over 1,200 elements nationwide. Examples of
these consumer and business related data elements include: Aggregated Home Values,
Average Consumer Age, Average Home Value, Percent White/Blue Collar, Percent
Family Households, Per Capita Income, Aggregated Income, Consumer Education Level,
Ethnicity Penetration, Employment Trends, Total Businesses, Industry/Business Types,
and hundreds more. MicroMarkets are predictive in defining revenue and/or customer
potential for every part of the country where population exists. By aggregating consumer
and business data, SourceLink MicroMarkets Analysis can rank and decile one or more
markets to drive comparisons and analysis on a local, regional, or national
level. MicroMarkets measure consumer and business information in distance bands from

LETS BE DIRECT

a center point to determine potential, competition and gravitational pull of each selected
market as it relates to a business. Analysis is aggregated to the 2.5, 5, 7.5 and 10 mile
radius around each MicroMarket.

The statistical segmentation methodology follows a three-phased process framework as


detailed below. SourceLink recommends using clustering methodology with K-Means
clustering technique, for the owner segmentation. Clustering is an undirected method
and does not require the selection of target variables. Clustering is preferred for the
following reasons:
o

o
o
o

Cluster analysis is based on data patterns to build segments that are distinct, i.e.,
homogenous within and heterogeneous between the segments. That is, all
similar stores showing the same behaviour would be clustered into one segment.
Where there are large numbers of independent explanatory variables, clustering
is a preferred technique: it allows for data reduction, providing information on
specific subgroups within a population.
Allows for the partition of the universe of TSC stores into similar clusters.
Allows for identification of relationships and profiling of segments within the store
universe.
Allows for better understanding of different types or groups of stores.

The K-Means cluster Analysis technique is preferred considering the large data sets and
the efficiency of the K-Means clustering technique with large data sets. It provides
superior results on new data sets in identifying the cluster seed points.

LETS BE DIRECT

Centroid
In the Discover phase, store and market profiling and segmentation are conducted using
a clustering technique. The clustering technique follows Exploratory Data Analysis,
applying a group of multivariate techniques to group objects based on the characteristics
that they possess. The objects are grouped in such a way that the degree of association
between two objects is maximal if they belong to the same group and minimal otherwise.
SourceLink ensures the robustness of the clustering output, by ensuring the best
practice guidelines for model effectiveness, as follows:

The clusters will be built on 100% of the store universe, so sampling bias is not
an issue.

Multicollinearity between variables should be low: Because each variable is


weighted equally in cluster analysis, more emphasis is placed on those variables
that are closely related, thus impacting interpretation of results. Hence variables
with heavy dependency/relationship are removed from the clustering variable set.

Outlier treatment: Outliers can distort data and make clusters unrepresentative of
the true population structure. Hence, we will remove the outliers for a separate
analysis to eliminate skew. Outlier cases are automatically segregated into their
own cluster in a two-stage cluster analysis.

SourceLink follows the interactive K-Means Clustering process for effective clustering
segmentation:

LETS BE DIRECT

Construction of initial cluster centers.


Assign cases to clusters based on distance from the cluster centers so as to
minimize variability within clusters and maximize variability between clusters.
Update the locations of cluster centers based on the mean values of cases in
each cluster.
Iterate until any reassignment of cases would make the clusters more internally
variable or externally similar.

 Kickoff Meeting (Pre-Analysis):


The primary objectives of the kickoff meeting are to clearly define and agree upon
business objectives, metrics of success, roles, and responsibilities. We will also review
the list of possible uses of the Segmentation and determine TSC variables SourceLink
will need for analysis.

 Segmentation Review (Post-Analysis):


SourceLink recommends a post-analysis review with TSC prior to finalizing the
segmentation. There will likely be some decisions to be made to choose between
alternative possible segmentations and combining segments to make them more
actionable. We will discuss various alternatives to ultimately determine the final
segmentation recommendations.

II.

Store Intercept Attribution Study

During our August 1st, 2012 meeting, TSC shared concerns about the current attribution
assumptions stating that the current database is directional, not exact. TSC would like
to understand how many and what types of households are actually missing from the
database and not represented in the customer universe they currently capture.
SourceLink recommends developing a store intercept program to capture 100%
attribution in 3 to 5 test stores for a period of approximately 14 days. Data from this test
period will be analyzed and compared to historical data to determine how accurate
current attribution assumptions are and provide invaluable database insight that can be
extrapolated to the entire database. SourceLink would work closely with TSC to identify
the test stores and develop the methodology to ensure the most representative results.

LETS BE DIRECT

This proposal includes SourceLinks consultation on the development of the study and
our back-end analysis of the intercept data overlaid with database data for gap analysis.
It does not include expenses related to execution of the study. We can provide a
comprehensive proposal that incorporates outsourcing the execution/capture if desired.
 Phase One: Pre Data Capture Initiative
Early in 2013, we can launch an attribution study to capture 100% of shoppers for a
limited time frame and geography to gain insights into the non-captured customer
universe. These insights can potentially shape the data capture initiative. We will
analyze the captured data in relation to the current TSC assumptions about attribution to
validate those assumptions. We can also execute this in Q1 to deliver insights quickly
to shape the 2013 strategies. In addition, having a more accurate understanding of
attribution can be used to set expectations and performance metrics for the planned
2013 threshold data capture initiative.
 Phase Two: Concurrent to Data Capture Initiative
We can periodically conduct the study to continue to validate assumptions with 100%
capture. We can also provide insights on customers where phone reverse append was
not a match. If the study is conducted periodically, we can also analyze trends over
time. We can structure it where part of the study is static (for trend analysis) and part is
custom to capture hot topics of interest.

III.

New Store Launch Acquisition Smart Model

LETS BE DIRECT

S.M.A.R.T. Modeling - SourceLinks Strategic Market Area Retail Targeting


(S.M.A.R.T.) Modeling technology is a proven statistical modeling system designed to
improve ROI in marketing promotions. S.M.A.R.T. can be used for customer retention,
cross sell or prospecting efforts. Unlike traditional modeling, S.M.A.R.T. includes
proprietary technology to model on a site by site basis. Whether store by store, region by
region or MSA or DMA, S.M.A.R.T. analyzes each location and can provide a limitless
number of custom models to cover the relevant geography.
SourceLinks proprietary S.M.A.R.T. Modeling is particularly relevant to TSCs goals for
new store grand openings. As we discussed during our August 1st meeting, we have
significant experience in leveraging our models against a global universe or a market by
market basis depending on TSCs goals and objectives. We will identify the best targets
for TSCs marketing efforts across the entire new store footprint. We can also use our
SUM methodology for response tracking to more accurately measure campaign results.
In 2011, SourceLink conducted a SMART model back test on TSC campaigns that
showed significant lift using a SMART model for acquisition targeting. The majority of
TSC responders were identified within the top deciles of the model. We can also
leverage our New Mover hygiene process to identify New Movers to include in an of our
acquisition or customer campaigns.

IV.

SourceLink Campaign Optimizer sm (Multivariate Testing)

SourceLinks Campaign Optimizer

sm

uses advanced testing techniques to

simultaneously test multiple variables of your marketing efforts at once while using the
same sample size as a standard split A/B test. We can test up to 8 variables
simultaneously for up to 32 test cells. Campaign Optimizer can be used to test mail
piece design, models, offers, and more. Not only will Campaign Optimizer provide
actionable insights on each variable tested, but it will also test for interactions between
the test elements, providing more detail than standard split A/B testing techniques. It is
a very cost effective way to produce significant response lift with a single mailing,
analysis, and deployment based on results.

LETS BE DIRECT

SourceLink can design multivariate tests to conduct and measure a number of tests
simultaneously (3 to 35). We can leverage campaign optimizer for direct mail, circulars,
e-mail and website tests. We can effectively test models, creative offers, design
elements, discounts, and price points within a single mailing.

V.

Secondary Customer Profiling

SourceLink will provide detailed customer profiles of TSC secondary customers.


Examples are:

younger generations

environmentally conscious customers

high end pet customers

convenience shoppers

We will profile them using the SourceLink National Consumer File. We can use both
those that we can identify on the TSC database leveraging transactional data as well as
prospects we can identify with similar characteristics and interests. We can then
compare and contrast the two profiles. This will help TSC leadership better understand
the demographics and needs of these potential new growth customers.

VI.

Custom Built Customer Segmentation Study

TSC has long used Mosaic Codes for customer segmentation. With the significant shift
in Mosaic Code structure, TSC is considering alternative segmentation schemas. While
off the shelf segmentations can serve a purpose, a custom built segmentation is often
much more relevant and actionable. We can also segment tax-exempt customers
separately and identify unique clusters such as small businesses and professional
farmers. We can segment the current database or the tax-exempt customers you
currently have captured, but the study would have much more depth and potential uses
if it were built on a sample following the store intercept project that was more reflective of
the entire customer base.

Investment Range
LETS BE DIRECT

I)

II)

III)

Description

Pricing Range

Sister Store Segmentation Schema 2 test pairs per segmentation

$20,000

Store Segmentation Roll Out with Profiles - i.e. Demographic, Competitive, &

$10,000 per

Product. We estimate 3 Segmentations.

Segmentation

New Store Assignments to all segmentations to be done in conjunction with

$1,200 per

master store updates (currently every 6 wks)

update

Quarterly Refresh of Existing Store Assignments

$5,000

Annual Evaluation and Segmentation Re-Build of three existing segmentations

$25,000

Ongoing Consultative Analytics

TBD

Store Intercept Attribution Study study can be quarterly or bi-annually with

$15,000

price adjusted according to the frequency.

$25,000

New Store Launch Acquisition Smart Model

$12,000 $15,000 $60M


scoring

IV)

Campaign Optimizer Multivariate Testing

$5,000-$16,000

V)

Secondary Customer Profile

$3,500 - $5,000
per customer
type

VI)

Custom Built Customer Segmentation Study (rollout with full profiles &

$30,000 -

presentation)

$45,000

LETS BE DIRECT